The relationship between working memory and

University of Wollongong

Research Online University of Wollongong Thesis Collection

University of Wollongong Thesis Collections

2003

The relationship between working memory and inhibition: the influence of working memory load on the interference and negative priming effects involved in selective attention Donna Bayliss University of Wollongong

Recommended Citation Bayliss, Donna, The relationship between working memory and inhibition: the influence of working memory load on the interference and negative priming effects involved in selective attention, Doctor of Philosophy thesis, Department of Psychology, University of Wollongong, 2003. http://ro.uow.edu.au/theses/1692

Research Online is the open access institutional repository for the University of Wollongong. For further information contact Manager Repository Services: [email protected].

The relationship between working m e m o r y and

inhibition: The influence of working memory load on the interference and negative priming effects involved in selective attention

A thesis submitted in fulfilment ofthe requirements for the award ofthe degree of

DOCTOR OF PHILOSOPHY

from

UNIVERSITY OF WOLLONGONG

By

Donna Bayliss BSc (Hons)

Department of Psychology 2003

Declaration

I, Donna M . Bayliss, declare that this thesis, submitted in fulfilment ofthe requirements

for the award of Doctor of Philosophy, in the Department of Psychology, University of Wollongong, is wholly my own work unless otherwise referenced or acknowledged. The document has not previously been submitted for qualifications at any other academic institution.

Donna M . Bayliss 28th April 2003

II

Dedication

I dedicate this thesis to m y husband, Gavin, w h o has been with m e the whole way.

His unwavering support and belief in me has meant more to me more than he will ever know. Thank you sweetheart.

Ill

Acknowledgments

There are a number of people who have contributed in their own way to my

development as a researcher and ultimately to the completion of this thesis. To ea them, I extend my sincere thanks. Firstly, to my supervisor, Steven Roodenrys, who first sparked my interest in cognitive psychology and has continued to provide me guidance and support throughout my undergraduate Honours thesis and Doctoral

research. Steve has assisted me in becoming an independent researcher, but most of he has taught me to believe in myself. To my parents, for their unconditional love

for understanding what it was that I was trying to achieve. Mum, you have always b there for me, and Dad, you still make me laugh. To my best friend Mel, for always

knowing exactly how I felt and exactly what to say to make it all seem better. You a fantastic scientific mind and an amazing way with people and you have been an

inspiration to me in both my research and my life. And finally, to Gav, who has lau

with me during the good times and suffered with me through the tough times. You ha given me endless support and love and I couldn't have done this without you.

IV

Abstract

The constructs of working m e m o r y and inhibition have been intimately linked in a number of cognitive theories. However, the exact nature of this relationship remains unclear. Roberts and Pennington (1996) proposed a framework in which the successful

inhibition of prepotent responses is a function of the strength of the prepotency, the working memory resources available to an individual and the working memory demands of the task. The prediction made from this framework is that increasing the working memory demands of a task will compromise working memory performance and decrease the ability to inhibit competing responses. This prediction was examined in the Stroop, flanker, and n-back paradigms by manipulating the working memory load of each task, and examining the effect on the inhibitory processes involved in the

interference and negative priming effects. In contrast to predictions, the interferen negative priming effects were unaffected by the introduction of a memory load, increasing the number of items in memory, maintaining a preload of memory items, or by actively maintaining and updating increasing numbers of items in working memory. However, the negative priming effect was eliminated when the working memory load involved a switch of attention away from the selective attention task. It was argued an interaction between working memory and inhibition will only become apparent when

the limited resources of the central executive are required. The implications of these results for models of working memory and selective attention are discussed.

V

Table of Contents Declaration II Dedication Ill Acknowledgments IV Abstract V Table of Contents VI List of Figures X Chapter 1 : Working Memory 1 1.1 Development ofthe working memory construct 1.2 Structure of working memory 1.3 Function of working memory 1.4 Characteristics of working memory 1.4.1 Capacity limitations 1.4.2 Control mechanisms 1.5 Measuring working memory 1.5.1 Dual-task performance 1.5.2 Complex span tasks 1.5.3 Structural equation modeling 1.6 Summary Chapter 2 : Inhibition and Interference in Cognition 2.1 Development ofthe constructs of inhibition and interference 2.1.1 Cognitive development and aging 2.1.2 Neuropsychology 2.1.3 Individual differences in working memory and general cognitive ability 2.2 Classification of Inhibition and Interference Processes 2.3 The relationship between interference and inhibition 2.4 Measuring inhibition and interference 2.4.1 Negative Priming 2.5 Mechanisms of inhibition 2.5.1 The Houghton-Tipper model 2.6 Summary Chapter 3 : Models of Working M e m o r y

1 5 9 13 13 17 19 19 21 24 25 27 27 27 30 33 36 40 47 47 55 56 59 61

3.1 The Multiple-Component Model (Baddeley & Hitch, 1974; Baddeley & Logie, 1999) 63 3.1.1 Structure 63 3.1.2 Function 67 3.1.3 Capacity limitations 68 3.1.4 Control mechanisms 70 3.1.5 Inhibitory processing 72 3.2 The Controlled Attention Model (Engle, Kane & Tuholski, 1999a) 73

VI

3.2.1 Structure 3.2.2 Function 3.2.3 Capacity limitations 3.2.4 Control mechanisms 3.2.5 Inhibitory processing 3.3 The Interactive Framework (Roberts & Pennington, 1996) 3.4 Overview of thesis Chapter 4 : The Stroop Task

73 75 77 78 79 83 86 91

4.1 Experiment 1 4.2 Method 4.2.1 Participants 4.2.2 Materials 4.2.3 Procedure 4.3 Results 4.3.1 Naming Latency 4.3.2 Recall Accuracy 4.3.3 Naming Errors 4.3.4 Naming Latency for Aware Participants 4.4 Discussion Chapter 5 : The Flanker Task

91 94 94 94 96 97 97 103 106 108 110 115

5.1 Experiment 2a 5.2 Method 5.2.1 Participants 5.2.2 Materials 5.2.3 Procedure 5.3 Results 5.3.1 Naming Latency 5.3.2 Recall Accuracy 5.3.3 Naming Errors 5.3.4 Naming Latency for Aware Participants 5.4 Discussion 5.5 Experiment 2b 5.6 Method 5.6.1 Participants 5.6.2 Materials 5.6.3 Procedure 5.7 Results 5.7.1 Naming Latency 5.7.2 Recall Accuracy 5.7.3 Naming Errors 5.7.4 Naming Latency for Aware Participants 5.8 Discussion 5.9 Experiment 2c 5.10 Method 5.10.1 Participants 5.10.2 Materials 5.10.3 Procedure 5.11 Results

VII

116 117 117 118 119 120 120 122 124 125 126 130 132 132 132 134 134 134 140 143 146 149 154 156 156 156 157 158

5.11.1 Naming Latency 5.11.2 Recall Accuracy 5.11.3 Naming Errors 5.11.4 Naming Latency for Aware Participants 5.12 Discussion 5.13 Experiment 2d 5.14 Method 5.14.1 Participants 5.14.2 Materials 5.14.3 Procedure 5.15 Results 5.15.1 Naming Latency 5.15.2 Recall Accuracy 5.15.3 Naming Errors 5.15.4 Naming Latency for Aware Participants 5.16 Discussion Chapter 6 : The N-Back Task

158 163 167 170 173 179 181 181 181 182 183 183 189 194 198 200 207

6.1 Experiment 3a 6.2 Method 6.2.1 Participants 6.2.2 Materials 6.2.3 Procedure 6.3 Results 6.3.1 Accuracy 6.3.2 Response Time 6.4 Discussion 6.5 Experiment 3b 6.6 Method 6.6.1 Participants 6.6.2 Materials 6.6.3 Procedure 6.7 Results 6.7.1 Accuracy 6.7.2 Response Time 6.8 Discussion

208 209 209 210 211 213 213 220 224 227 228 228 229 229 230 230 236 240

Chapter 7 : Summary and Discussion

245

7.1 Review of empirical evidence 7.2 Theoretical implications for models of working memory 7.2.1 The Interactive Framework (Roberts & Pennington, 1996) 7.2.2 The Multiple-Component Model (Baddeley & Logie, 1999) 7.2.3 The Controlled Attention Model (Engle, Kane & Tuholski, 1999a) 7.3 Theoretical implications for mechanisms of negative priming 7.3.1 Selective Inhibition 7.3.2 Episodic Retrieval 7.3.3 Tipper's view of negative priming 7.4 Directions for future research 7.5 Conclusions References

246 255 256 260 264 267 267 271 274 276 281 285

VIII

Appendix A

307

Appendix B 312 Appendix C 331

IX

List of Figures Figure 3.1: Diagram adapted from Engle, Conway, Tuholski and Shisler (1995) showing the procedure used in their letter-naming task

80

Figure 4.1: Naming latency in the Stroop task with and without a m e m o r y load as a function of condition and list length, collapsed across serial position

98

Figure 4.2: Naming latency in the Stroop task with a 4-item m e m o r y load as a function of condition and serial position

100

Figure 4.3: Naming latency in the Stroop task with a 5-item m e m o r y load as a function of condition and serial position

101

Figure 4.4: Naming latency in the Stroop task with a 6-item m e m o r y load as a function of condition and serial position

102

Figure 4.5: Recall accuracy in the 4, 5, and 6-item recall tasks as a function of condition, collapsed across serial position

103

Figure 4.6: Recall accuracy in the 4-item recall task as a function of condition and serial position

104

Figure 4.7: Recall accuracy in the 5-item recall task as a function of condition and serial position

105

Figure 4.8: Recall accuracy in the 6-item recall task as a function of condition and serial position

106

Figure 4.9: Naming errors in the Stroop task with and without a m e m o r y load as a function of condition and list length

107

Figure 4.10: Naming latency of aware participants in the Stroop task with and without a m e m o r y load as a function of condition and list length, collapsed across serial position

109

Figure 5.1: Naming latency in the naming and recall tasks as a function of condition and serial position

121

Figure 5.2: Recall accuracy in the colour recall task as a function of condition and serial position

123

Figure 5.3: Naming errors in the naming task and recall task as a function of condition 124 Figure 5.4: Naming latency of aware participants in the naming task and recall task as a function of condition and serial position

X

125

Figure 5.5: N a m i n g latency in the naming task as a function of condition and serial position

135

Figure 5.6: N a m i n g latency in the letter recall task as a function of condition and serial position

135

Figure 5.7: N a m i n g latency in the colour recall task as a function of condition and serial position

136

Figure 5.8: N a m i n g latencies of low-span participants as a function of condition and serial position collapsed across task

138

Figure 5.9: N a m i n g latencies of high-span participants as a function of condition and serial position collapsed across task

139

Figure 5.10: Recall accuracy in the letter recall and colour recall tasks as a function of condition and serial position

140

Figure 5.11: Recall accuracy of low-span participants in the letter recall and colour recall tasks as a function of condition and serial position

142

Figure 5.12: Recall accuracy of high-span participants in the letter recall and colour recall tasks as a function of condition and serial position

142

Figure 5.13: N a m i n g errors in the naming, letter recall and colour recall tasks as a function of condition

144

Figure 5.14: N a m i n g errors in the naming, letter recall and colour recall tasks as a function of condition and span group

145

Figure 5.15: N a m i n g latency of aware participants in the naming task as a function of condition and serial position

146

Figure 5.16: N a m i n g latency of aware participants in the letter recall task as a function of condition and serial position

147

Figure 5.17: N a m i n g latency of aware participants in the colour recall task as a function of condition and serial position

147

Figure 5.18: N a m i n g latency in the naming task as a function of condition and serial position

158

Figure 5.19: N a m i n g latency in the 1-item preload task as a function of condition and serial position

159

Figure 5.20: N a m i n g latency in the 4-item preload task as a function of condition and serial position

159

XI

Figure 5.21: N a m i n g latencies of low-span participants as a function of condition and serial position collapsed across task

162

Figure 5.22: N a m i n g latencies of high-span participants as a function of condition and serial position collapsed across task

162

Figure 5.23: M e a n proportion of colours correctly recalled in the 4-item preload task as a function of serial position and condition

164

Figure 5.24: Recall accuracy of low-span participants in the 4-item preload task as a function of condition and serial position

166

Figure 5.25: Recall accuracy of high-span participants in the 4-item preload task as a function of condition and serial position

166

Figure 5.26: N a m i n g errors in the naming, 1-item preload and 4-item preload tasks as a function of condition

168

Figure 5.27: N a m i n g errors in the naming, 1-item preload and 4-item preload tasks as a function of condition and span group

169

Figure 5.28: N a m i n g latency of aware participants in the naming task as a function of condition and serial position

170

Figure 5.29: N a m i n g latency of aware participants in the 1-item preload task as a function of condition and serial position

171

Figure 5.30: N a m i n g latency of aware participants in the 4-item preload task as a function of condition and serial position

171

Figure 5.31: N a m i n g latency in the unmasked naming and colour recall tasks as a function of condition and serial position

183

Figure 5.32: N a m i n g latency in the masked naming and colour recall tasks as a function of condition and serial position

184

Figure 5.33: N a m i n g latencies of low-span participants as a function of task, condition and serial position in the unmasked condition

186

Figure 5.34: N a m i n g latencies of low-span participants as a function of task, condition and serial position in the masked condition

187

Figure 5.35: N a m i n g latencies of high-span participants as a function of task, condition and serial position in the unmasked condition

187

Figure 5.36: N a m i n g latencies of high-span participants as a function of task, condition and serial position in the masked condition

XII

188

Figure 5.37: M e a n proportion of colours correctly recalled in the unmasked colour recall task as a function of serial position and condition

190

Figure 5.38: M e a n proportion of colours correctly recalled in the masked colour recall task as a function of serial position and condition

190

Figure 5.39: Recall accuracy of low-span participants in the unmasked recall task as a function of condition and serial position

192

Figure 5.40: Recall accuracy of low-span participants in the masked recall task as a function of condition and serial position

192

Figure 5.41: Recall accuracy of high-span participants in the unmasked recall task as a function of condition and serial position

193

Figure 5.42: Recall accuracy of high-span participants in the masked recall task as a function of condition and serial position

193

Figure 5.43: N a m i n g errors in the naming and colour recall tasks as a function of masking and condition

195

Figure 5.44: N a m i n g errors in the naming and colour recall tasks as a function of condition and span group

197

Figure 5.45: N a m i n g latency of aware participants in the unmasked naming and colour recall tasks as a function of condition and serial position

198

Figure 5.46: N a m i n g latency of aware participants in the masked naming and colour recall tasks as a function of condition and serial position

199

Figure 6.1: M e a n number of correct responses in the single letter and letter triplet conditions ofthe n-back task as a function ofthe level of working m e m o r y load. 214 Figure 6.2: M e a n number of correct responses for the low-span participants as a function of interference condition and level of working m e m o r y load

215

Figure 6.3: M e a n number of correct responses for the high-span participants as a function of interference condition and level of working m e m o r y load

216

Figure 6.4: M e a n d' values in the single letter and letter triplet conditions ofthe n-back task as a function of working m e m o r y load

217

Figure 6.5: M e a n d' values for low-span participants in the single letter and letter triplet conditions ofthe n-back task as a function of working m e m o r y load

218

Figure 6.6: M e a n d' values for high-span participants in the single letter and letter triplet conditions ofthe n-back task as a function of working m e m o r y load XIII

219

Figure 6.7: M e a n response times in the single letter and letter triplet conditions ofthe nbacktask as a function ofthe level of working m e m o r y load

221

Figure 6.8: M e a n response times ofthe low-span participants in the single letter and letter triplet conditions ofthe n-back task as a function ofthe level of working m e m o r y load

222

Figure 6.9: M e a n response times ofthe high-span participants in the single letter and letter triplet conditions ofthe n-back task as a function ofthe level of working m e m o r y load

223

Figure 6.10: M e a n number of correct responses in the single letter and letter triplet conditions ofthe n-back task as a function ofthe level of working m e m o r y load. 230 Figure 6.11: M e a n number of correct responses for the low-span participants as a function of interference condition and level of working m e m o r y load

232

Figure 6.12: M e a n number of correct responses for the high-span participants as a function of interference condition and level of working m e m o r y load

232

Figure 6.13: M e a n d' values in the single letter and letter triplet conditions ofthe n-back task as a function of working m e m o r y load

234

Figure 6.14: M e a n d' values for low-span participants in the single letter and letter triplet conditions ofthe n-back task as a function of working m e m o r y load

235

Figure 6.15: M e a n d' values for high-span participants in the single letter and letter triplet conditions ofthe n-back task as a function of working m e m o r y load

236

Figure 6.16: M e a n response times in the single letter and letter triplet conditions ofthe n-back task as a function ofthe level of working m e m o r y load

237

Figure 6.17: M e a n response times ofthe low-span participants in the single letter and letter triplet conditions ofthe n-back task as a function ofthe level of working m e m o r y load

238

Figure 6.18: M e a n response times ofthe high-span participants in the single letter and letter triplet conditions ofthe n-back task as a function ofthe level of working m e m o r y load

239

XIV

Chapter 1 : Working Memory 1.1 Development ofthe working memory construct

The term "working memory" was first used by Miller, Galanter, and Pribram (1960) to describe a quick-access system responsible for the storage of transient

information required for the execution of current behavioural strategies, referred to Miller and colleagues as "plans". The contents of this system were thought to be immediately accessible to conscious awareness and separate from the more usual

storage of information now referred to as long-term memory. The distinction between a

short-term transient store and a more permanent form of memory was also central to th model of Waugh and Norman (1965). They refer to primary memory as being a limited capacity store of current information and secondary memory as a larger, more durable store. The information in primary memory is continually replaced by new material unless it is rehearsed; in which case the information may then be transferred into secondary memory (Waugh & Norman, 1965). This terminology was originally proposed by William James (1890), who described primary memory as the existing events in consciousness, while secondary memory referred to the recollection of past

events. The critical difference between the ideas of James and the models ofthe 1960s is that the later models implicitly incorporated higher-order functions that enabled manipulation ofthe short-term information in addition to storage. This provided the basis for contemporary models of working memory. Atkinson and Shiffrin (1968) explicitly described the involvement of control processes in their influential model of human memory. They regarded the short-term store component of their model as the control centre for the whole memory system,

controlling the flow of information between the sensory register, the long-term store 1

response output mechanisms (Atkinson & Shiffrin, 1968). A number of control

processes necessary for the task of "remembering" were ascribed to the short-term s

including coding and rehearsal, decision making and retrieval strategies (Atkinson & Shiffrin, 1971). The Atkinson and Shiffrin framework in a sense "raised the bar" in working memory research, not only by specifying the control processes thought to be involved, but by assigning a central role to working memory in the information processing system. In one ofthe most influential papers in the working memory literature, Baddeley and Hitch (1974) incorporated the notion of a short-term store into a more complex

system thought to be essential to everyday cognitive activities. In contrast to Atk and Shiffrin's (1968) concept of a short-term store, Baddeley and Hitch (1974) proposed a multi-component model, which distinguished between a limited capacity

storage component dedicated to the rehearsal of speech-based information and a cent processing system. This model was, in part, based on results from a series of experiments examining the effect of various memory loads on reasoning, comprehension and free recall performance. The memory load manipulation involved

the presentation of a list of items, either prior to or concurrently with the prese

the primary task, which were to be recalled in serial order at the end of each trial found that while a concurrent memory load of six items impaired performance in all three tasks (i.e. reasoning, comprehension and free recall), a memory load of up to items produced a negligible effect. This was taken as evidence that the maintenance

list of items could be performed by the storage component independently ofthe centra processing space, provided the memory load was within the limited capacity ofthe storage component. If this capacity was exceeded, the central component was thought assist with storage resulting in impaired performance on the primary task (Baddeley 2

Hitch, 1974). Baddeley and Hitch (1974) believed the storage component to be

phonemically based, describing it as a "phonemic response buffer which is able to store a limited amount of speech-like material in the appropriate serial order" (p.77). They also suggested that a similar component existed for visual memory, which was responsible for the short-term storage of visual information. Evidence for the separability ofthe two storage systems came from a study by Brooks (1967, Experiment 1). He demonstrated that performance on a memory task requiring the recall of a series of sentences that described various spatial relations and were therefore amenable to visual imagery was better with auditory presentation ofthe sentences than when the

sentences were concurrently read from a written passage. In contrast, performance on an equivalent verbal memory task requiring the recall of nonsense sentences that did not describe spatial relations was better when the sentences were presented visually in a written passage than when they were presented auditorily. Brooks (1967) argued that

reading the sentences required the use ofthe visual system and therefore interfered wi the generation of a visual representation ofthe spatial relations described in the sentences, whereas listening to the sentences did not. This provided preliminary evidence ofthe dissociation ofthe verbal and visual short-term storage systems. The central processing component ofthe Baddeley and Hitch model was loosely described as "a more flexible and executive component" (Baddeley & Hitch, 1974, p.77) thought to be responsible for encoding and retrieval strategies as well as any control-processing functions. Baddeley and Hitch (1974) did not attempt to specify the

central component of their original model in any detail, but alluded to the crucial ro was likely to play in the working memory system. In a later description ofthe model, Baddeley (1986) incorporated the Norman and Shallice (1986) model of attentional

control as a possible account of central executive functioning. This framework include 3

a limited capacity supervisory attentional system ( S A S ) responsible for the conscious control of behaviour in tasks or situations requiring planning, problem solving, novel actions and the overcoming of habitual responses when these responses are no longer appropriate (Norman & Shallice, 1986). This encouraged the conceptualisation ofthe central executive component of working memory as a limited attentional resource1. In a restatement of his original position, Baddeley (1993) described the central executive component as "concerned with attention and co-ordination rather than storage" (p. 168) and acknowledged that tasks involving working memory did not necessarily involve memory per se. However, Baddeley (1993) was clear in his belief that temporary storage was "an absolutely essential feature ofthe working memory system as a whole" (p. 168). Thus, Baddeley and his colleagues (Baddeley, 1986; Baddeley, 1993; Baddeley & Hitch, 1974) extended the functions of working memory beyond simple storage to incorporate control mechanisms responsible for the active processing of information and coordination ofthe system as a whole. The model of Baddeley and Hitch (1974) was instrumental in shaping the current conceptualisation of working memory. In contrast to earlier models of a shortterm store, greater emphasis was placed on the functional role of working memory in

1

The terms 'resource' and 'capacity' are often used interchangeably in the literature wi

consideration of what is meant by these terms. 'Capacity' refers to the limits of a part

as the amount of information that can be stored, whereas 'resource' refers to the source

attention that is drawn upon in order to perform various cognitive activities. Confusion

capacity of a system is often determined by the availability of a resource. However, the

system and the resources the system draws upon are not necessarily the same thing. For ex Kurland and Goldberg (1982) argued that increases in memory capacity are not the result

attentional resources per se, but an increase in operational efficiency that frees up av storage. 4

complex cognitive activities. This relationship was the focus of a seminal study by Daneman and Carpenter (1980) in which they developed one ofthe first measures of

working memory capacity, which they called the reading span task (see 1.3 Function working memory). Daneman and Carpenter (1980) showed that performance on this

task was closely related to measures of reading comprehension, whereas performance on a test of short-term memory ability was not. An extensive body of research has followed in this tradition of examining individual differences in working memory

capacity and their relationship to higher-level cognitive abilities. The concept o working memory is now used to refer to an integrated system responsible for the

temporary maintenance and active manipulation of task relevant information (Badde 2000b; Becker & Morris, 1999; Miyake & Shah, 1999a). In spite ofthe general acceptance and widespread use ofthe term, agreement has not been reached on the fundamentals ofthe working memory system. As a result, there are a number of

seemingly disparate theories of working memory, which, not surprisingly, has led t variety of different approaches to measuring working memory performance. This was illustrated in a recent book edited by Miyake and Shah (1999b) that provided a description and comparison of a number ofthe current models of working memory

based on a specific set of theoretical questions. Adopting their approach, the fo review will attempt to summarise the working memory literature in terms ofthe

structure, function and general characteristics of working memory, as well as comm methods of measuring working memory performance.

1.2 Structure of working memory

One ofthe main areas of contention regarding the structure of working memory has been whether working memory comprises a unitary or non-unitary system. The 5

multiple-component model of Baddeley and Hitch (1974) has been influential in

advocating the non-unitary nature of working memory with domain-specific subsystem dedicated to verbal and visuospatial processing and storage. Evidence for the

separability ofthe visual and verbal subsystems is largely based on dual task stu

demonstrate double dissociations between performance on visual and verbal tasks wh

combined with secondary tasks that are also visual and verbal in nature (Logie, Zu & Baddeley, 1990). Other proponents ofthe non-unitary view have divided working

memory on the basis ofthe codes or representations involved (Schneider & Detweiler

1987), or the brain areas implicated with different types of processing (Awh et al Jonides et al., 1998a; Smith & Jonides, 1999). In contrast, researchers favouring a unitary view of working memory have

tended to focus on the domain-general aspects of working memory. For example, Engl Kane and Tuholski (1999a) describe their model of working memory as a system

consisting ofthe long-term memory traces active above threshold, the procedures an skills for maintaining activation, and controlled attention, which they liken to executive component of Baddeley and Hitch's model. However, Engle and his colleagues have mostly focused on the limited capacity ofthe controlled attention component, which they believe is domain-free and represents a unitary working

memory/attention system (Engle et al., 1999a). Support for this view is provided b number of studies in which measures of working memory capacity have been found to

predict performance on a range of complex cognitive activities (Daneman & Carpente 1980; Daneman & Merikle, 1996; Kyllonen & Christal, 1990; Tuholski, Engle, &

Bay lis, 2001). In addition, a recent latent-variable study from the Engle lab sho a number of working memory tasks loaded significantly on a single working memory

factor, which was separable from a second factor representative of short-term memo 6

(Engle, Tuholski, Laughlin, & C o n w a y , 1999b). However, Engle et al. (1999a) do not deny the existence of domain-specific codes and suggest that working memory may form a hierarchical structure with a general domain-free system controlling a number

domain-specific subsystems. In this respect, the unitary model of Engle et al. (1999a does not appear to be fundamentally different from the non-unitary models that incorporate separate components to account for domain-specific effects such as the interference of articulatory suppression and spatial tapping with verbal and visualspatial information respectively (Baddeley, 1986). Even Cowan (1999), who does not

distinguish between different modalities or representations in his concept of activat

memory, acknowledges that his model is not fully unitary as it includes both a passiv storage and active processing component. Indeed, as Miyake and Shah (1999c) propose, a more useful way of differentiating current models of working memory may be in

terms ofthe domain-specific and domain-general factors they incorporate and the exten to which each contributes to working memory performance. Another related but separate issue concerning the structure of working memory is whether working memory and long-term memory are structurally separable systems. Early short-term store models proposed a structural distinction between short-term storage and the more durable storage attributed to long-term memory (Atkinson & Shiffrin, 1971; Waugh & Norman, 1965) . This was to account for the primacy and

2

Although Atkinson and Shiffrin (1971) maintain that their account ofthe short-term store may be

considered as the temporary activation of long-term memory, they also equate the short

"consciousness" which leads me to agree with the view of Cowan (1993) that these two d

be used interchangeably. This is because of implicit priming effects, which demonstrat

memory outside of awareness. Moreover, Atkinson and Shiffrin (1971) clearly describe t

information from one store to another, which seems more compatible with a distinct sys 7

recency effects typically found infreerecall tasks. The recency effect was thought to

reflect the recall ofthe most recent items from the short-term store whereas earli

in the list were thought to benefit from extra rehearsal, which led to their trans the more permanent long-term store (Atkinson & Shiffrin, 1971; Waugh & Norman,

1965). One implication of this structurally distinct view of memory was that the sh term store was considered to be the mechanism responsible for the transfer of information to the long-term store. This view was challenged by data from a

neuropsychological patient K.F., who was reported as having normal long-term learn

abilities despite a severely impaired verbal short-term memory (Shallice & Warring 1970). However, Baddeley and Hitch (1974) believed that an impaired phonemic storage component but intact central executive component of their working memory

model could explain the performance of K.F. and other patients with similar deficit An alternative to the structural view depicts working memory as an activated

subset of long-term memory (Cowan, 1988; Engle et al., 1999a). This view has become

attractive in the face of mounting evidence for the contribution of long-term knowl

such as lexical information (Hulme, Maughan, & Brown, 1991; Hulme et al., 1997), to short-term memory performance. According to Cowan (1988), working memory

involves both the subset of long-term memory that is activated but outside consciou

awareness and the subset of activated memory that is currently in the focus of att In a comparable model, Engle et al. (1999a) described working memory as the long-

term traces active above threshold together with controlled attention. These models propose a more continuous view ofthe relationship between working memory and longterm memory than the structural models described above. However, it seems that the models are beginning to converge on this issue in that even those that propose a distinction between working memory and long-term memory do not necessarily believe 8

that these two systems are structurally distinct. For example, although Baddeley and Logie (1999) recently reiterated their view that working memory and long-term memory were functionally distinct systems, they hypothesised that the two systems may correspond to a neural network with fast and slow weights. Thus according to most of these models, the major distinction between working memory and long-term memory appears to be functional rather than structural.

1.3 Function of working memory

The working memory construct was originally proposed as a description ofthe type of temporary memory thought to be required in the performance of complex cognitive activities such as reading comprehension, learning and reasoning (Baddeley Hitch, 1974). These activities typically require the maintenance of task-relevant

information in order to facilitate the processing and manipulation of this informatio which is necessary for successful task performance. The functional role of working memory has been most clearly demonstrated in the study of reading and language comprehension. Daneman and Carpenter (1980) believed that working memory was required in reading comprehension to store the semantic and syntactic information necessary to interpret and integrate successive text. They argued that individual

differences in reading comprehension reflected differences in the functional capacit working memory, that is, the capacity for simultaneous storage and processing. They

argued that individuals with inefficient reading processes would need to allocate mor of their available resources to the processing requirements ofthe task leaving less

capacity for storage, and hence, difficulties with comprehension. In order to assess functional capacity of working memory, Daneman and Carpenter (1980) developed a reading span task that placed simultaneous demands on the storage and processing 9

functions of working m e m o r y (see Complex span tasks below). Performance on the

reading span task was found to correlate significantly better than a simple word spa test with two measures of comprehension involving fact retrieval and pronominal reference (Daneman & Carpenter, 1980). In addition, the reading span test was found

correlate with a global test of language comprehension (the Verbal Scholastic Aptitu Test, or VSAT) whereas the word span test did not. Daneman and Carpenter (1980) argued that individual differences in working memory capacity, as measured by the reading span test, were a crucial source of individual differences in language comprehension. In a later study, Daneman and Carpenter (1983) examined the role of working

memory capacity in the ability to integrate ambiguous information within and between sentences. They found that readers with smaller working memory spans had difficulty interpreting garden path sentences and were less able to integrate disambiguating information than readers with larger working memory spans. This was particularly evident when the disambiguating information was provided in a separate sentence, suggesting that readers with smaller working memory spans were more likely to lose

information across sentence boundaries. Similarly, Just and Carpenter (1992) found t individuals with larger working memory capacities were able to maintain multiple

interpretations of ambiguous words and so, were more likely than low span individual to have the correct interpretation available when required. However, this came at a

in terms of increased reading times for ambiguous sentences (Just & Carpenter, 1992) Thus, working memory plays a critical role in maintaining task relevant information enables readers to compute semantic and syntactic relations among words and

sentences, integrate text and ideas, retrieve facts, resolve ambiguities and make th inferences necessary for successful language comprehension. 10

The importance of working m e m o r y has been demonstrated in a number of other complex cognitive activities. Gilhooly, Logie, Wetherick and Wynn (1993) found that syllogistic reasoning performance was disrupted by concurrent random number generation but not by articulatory suppression or spatial tapping, which was taken as evidence that the strategies used to solve syllogistic reasoning tasks place demands central executive component of working memory. However, the exact nature ofthe working memory involvement in syllogistic reasoning was not established (Gilhooly et al., 1993). Reber and Kotovsky (1997) examined the role of working memory in problem-solving using a task similar to the Tower of Hanoi task and found that a secondary working memory load impaired problem-solving performance proportionate

to the level of working memory load imposed. However, the interesting finding was that a working memory load did not impair problem-solving performance the second time the problem was solved. Furthermore, Reber and Kotovsky (1997) assessed

participants' knowledge ofthe puzzle solution and found that very few were able to giv a verbal account ofthe solution strategy despite improved performance on the second

attempt, which led them to conclude that participants were learning to solve the puzzl

implicitly. The effect of working memory was limited to first trial performance, which

suggested that it was the implicit learning of strategies to solve the puzzle, but no execution of these strategies, that was dependent on working memory (Reber &

Kotovsky, 1997). The role of implicit learning in problem-solving tasks is not altoget surprising, however, the relationship of implicit learning to working memory capacity may prove to be crucial to our understanding ofthe acquisition of complex cognitive skills. In recent years, the relationship between working memory and the ability to

resist interference from task-irrelevant information has gained considerable attentio 11

3 0009 03299328

from a number of working m e m o r y theorists. Engle and his colleagues have argued that

individual differences in working memory capacity reflect differences in the capabil

for controlled attention in the face of distraction from the environment or interfer

from information that is competing for limited attentional resources (Conway & Engle, 1994; Engle, 1996; Engle et al., 1999a; Rosen & Engle, 1998). A series of studies conducted by Engle and his colleagues has demonstrated a relationship between

working memory capacity and the ability to resist interference from previously retri exemplars during a verbal fluency task (Rosen & Engle, 1997), response competition from temporarily activated but irrelevant target information (Conway & Engle, 1994), and the ability to suppress intrusions from previously learned information (Rosen &

Engle, 1998). In addition, Kane and Engle (2000) recently found a relationship betwee

working memory capacity and the ability to use controlled attention to reduce the ef of proactive interference from previously memorised information. A number of other

researchers have suggested that differences in the efficiency of inhibitory processi working memory is a major factor in cognitive development (Bjorklund & Harnishfeger, 1990; Harnishfeger & Bjorklund, 1993), cognitive aging (Hasher,

Stoltzfus, Zacks, & Rypma, 1991; Hasher & Zacks, 1988), and individual differences in

general cognitive ability (Dempster & Corkill, 1999). This issue will be discussed in

greater detail in a later section (2.1.1 Cognitive development and aging); however, i

evident that the ability to control interference is related to differences in working memory capacity, although the exact nature ofthe relationship is still a contentious issue.

12

1.4 Characteristics of working memory

Two characteristics of working memory typically referred to in most working

memory models are the limitations that constrain the capacity of working memory and the mechanisms that control and regulate the flow of information both into and out working memory.

1.4.1 Capacity limitations

Although it is generally accepted that working memory is limited in capacity, the nature ofthe constraints responsible for those limitations remains the subject considerable debate. Examinations of individual differences in working memory

capacity assume that there is considerable individual variation in the ability to a

maintain information for use in cognitive tasks. The specific capacity-limiting fact

proposed to account for this variation are diverse and provide one way of distingui between working memory models. One view is that the capacity of working memory is

a reflection ofthe maximum amount of activation available to support working memory functions. According to Just and Carpenter (1992) working memory consists of those

representational elements that are activated above some minimum threshold. Individu differ in the total amount of activation they have available for maintaining these

elements and differences in performance on working memory capacity tasks are though

to reflect this variation (Just & Carpenter, 1992). Just and Carpenter (1992) simula

this theory using a hybrid production/activation-based connectionist system and wer

able to demonstrate that limiting the amount of activation the system had available propagate activation to other elements resulted in similar performance to human

participants in several aspects of language comprehension. Although Just and Carpen

(1992) restricted their model to language comprehension, they believed that activat 13

limitations could be responsible for individual differences in other cognitive domains as well. Cantor and Engle (1993) examined the relationship between working memory capacity and long-term memory activation limits using a fan effect procedure. They found that as the number of concepts shared by different sentences increased, low

capacity participants showed a larger increase in verification time for target sente than high capacity participants. This was taken as evidence that low capacity

participants have less activation to distribute among the associated concepts and so, concepts in a larger association network receive less activation and take longer to the threshold necessary for recognition (Cantor & Engle, 1993). However, Conway and Engle (1994) found that response times in a memory search task were dependent on working memory capacity only if the target items were used in more than one memory

set, thus creating a degree of interference. These results could not be accounted for terms of activation limitations, which led Conway and Engle (1994) to propose a new view that attributed differences in working memory capacity to differences in attentional resources necessary for the inhibition of task-irrelevant information. According to this view, inhibition is resource demanding and as a consequence, individuals with greater attentional resources will also have a greater capacity for

preventing irrelevant information from interfering with the contents of working memo (Conway & Engle, 1994). Thus, limitations in working memory capacity will only be evident in situations that demand controlled attention. The efficiency of inhibitory processing has been proposed as another constraint on working memory performance. Hasher and Zacks (1988) argued that inhibitory mechanisms serve to restrict irrelevant information from entering working memory as

well as suppressing active information no longer relevant to the task at hand. Ineff 14

inhibitory mechanisms would result in more irrelevant information entering and

interfering with the active contents of working memory. In contrast to theories that claim the capacity of working memory determines the ability to handle interference, framework proposes that inhibitory mechanisms are responsible for the efficient

operation of working memory. Hasher and Zacks (1988) argued that age-related deficit

in working memory were a result of declining inhibitory efficiency. In support of th view, Hasher and Zacks and their colleagues have provided considerable evidence

demonstrating that older adults have difficulty suppressing extraneous information a

are more likely to maintain irrelevant information than younger adults (Hasher, Quig May, 1997; Hasher et al., 1991; Hasher & Zacks, 1988; Stoltzfus, Hasher, & Zacks,

1996) (see 2.1.1. Cognitive development and aging). Although Hasher and Zacks (1988)

developed their view as an alternative to capacity theories, it is difficult to dete

empirically whether performance limitations are due to interference between relevant

and irrelevant information concurrently active in working memory or if the maintenan

of irrelevant information results in less capacity for the storage and processing of

relevant information. More recently, Stoltzfus, Hasher and Zacks (1996) have suggest

that a synthesis ofthe inhibition and capacity view of working memory limitations ma be beneficial. Another limiting factor that has been popular as an explanatory variable in developmental and aging research is processing efficiency. Case, Kurland, and Goldberg (1982) argued that developmental increases in memory span are not due to

increases in total processing space, but increases in operational efficiency. Accord

Case et al. (1982) mental operations become faster and more efficient with developme

and as a result, the operations place less demands on overall resources leaving more

available to meet storage requirements. Case et al. (1982) demonstrated that limitin 15

processing speed of adults to that of a six-year-old child led to similar performance by

both groups on a test of working memory capacity. This finding provides support for th theory that working memory capacity is constrained by processing efficiency. At the opposite end ofthe lifespan, Salthouse and his colleagues (Salthouse, 1996; Salthouse & Babcock, 1991) have argued that an age-related decline in processing speed is a major contributor to the decline in cognitive performance observed in older adults. Salthouse (1996) argued that slower processing prevents cognitive operations from

being completed within the available time and reduces the amount of information that i simultaneously available when it is needed for later processing. In support of this theory, Salthouse provides considerable evidence that age-related influences on cognitive measures are significantly reduced once processing speed is statistically removed (Salthouse, 1996; Salthouse & Babcock, 1991; Salthouse & Meinz, 1995). The different theories discussed here suggest that there may be a number of factors that constrain working memory performance. Indeed, a recent review of current working memory theories led Miyake and Shah (1999c) to proclaim the demise ofthe assumption that working memory limitations arise from a single mechanism. Thus, a more systematic investigation ofthe relative importance of different capacityconstraining factors for specific tasks and populations may be advantageous. Woltz

(1988) used this approach to investigate the role of working memory in procedural skil acquisition. He distinguished between two sources of working memory limitations, namely controlled attention and automatic activation, and examined the influence of

these limitations over the course of a procedural learning task. The results demonstr that measures of controlled attention predicted early rule acquisition and proceduralisation, whereas measures of automatic activation predicted the later composition and strengthening of initial productions. Woltz (1988) concluded that the 16

two working m e m o r y constructs were distinct and imposed unique limits on the

processes of skill acquisition. Although this approach may prove to be difficult to to performance on other cognitive tasks, these results highlight the importance of understanding the specific constraints that contribute to individual differences in working memory performance.

1.4.2 Control mechanisms

In recent years, the mechanisms implicated in the control and regulation of

working memory have been the subject of considerable interest. Most models propose a central control structure, such as the central executive (Baddeley & Hitch, 1974;

Cowan, 1988; Engle et al., 1999a), to explain the control functions of working memor However, this approach has been criticised for creating a homunculus to account for

those aspects of performance that cannot be attributed to the operation of slave sy

or levels of automatic activation. Recent attempts to address this criticism have l

greater specification ofthe functions attributed to the control mechanism, however,

detailed account of how these functions relate to one another is required to overcom the homunculus problem. Some ofthe executive functions commonly referred to include the inhibition of prepotent or strongly triggered responses (Bayliss & Roodenrys, 2000; Roberts, Hager, & Heron, 1994), the capacity to switch attention between task sets (Allport, Styles, & Hsieh, 1994; Rogers & Monsell, 1995), the dynamic updating and manipulation ofthe contents of working memory (Jonides et al., 1997; Morris & Jones, 1990), and the strategic planning involved in problem solving

(Shallice, 1982). A recent latent variable analysis by Miyake et al. (2000) demonstr

the separability of three of these executive functions: task-switching, memory upda and inhibition (see 1.5.3 Structural equation modeling, for more detail). However, 17

having said that, the three factors representing these executive functions were

significantly correlated with each other which suggests that there may be a second o

factor that is common to all three. Thus, the question of whether these functions are independent of one another or are mediated by a smaller number of underlying control processes such as activation or inhibition remains to be seen. Alternatives to the central executive approach have been drawn from computational models and comparative research. Kimberg and Farah (1993) proposed a unified account ofthe cognitive impairments observed in frontal lobe patients on a range of seemingly disparate executive functions tasks. They demonstrated that weakening the strength among working memory associations in a production system

model produced failures that resembled the performance of frontal lobe patients on f executive function tasks. This was taken as evidence that a single underlying

impairment could account for a range of executive function deficits without resortin a central executive to account for complex cognitive performance (Kimberg & Farah, 1993). Based on extensive research with nonhuman primates, Goldman-Rakic (1995) argued that working memory functions could be explained by multiple domain-specific

systems organised in parallel rather than a central domain-general executive process Neurological examinations ofthe prefrontal cortex of nonhuman primates suggest that specialised domains have the neural complexity to register, maintain and process information by interacting with relevant sensory and motor areas (Goldman-Rakic,

1995). Although it may be argued that the human brain is architecturally different fr the nonhuman primate brain, Goldman-Rakic's (1995) model of working memory

suggests that executive control may be an emergent feature of a number of interactive neural systems. Continued development of both computational and comparative models

18

of executive functioning m a y prove to be beneficial in the continued attempt to eliminate the homunculus from working memory.

1.5 Measuring working memory As the more traditional concept of a short-term store became incorporated within the more complex conceptualisation of working memory, it became apparent that

conventional measures of memory span failed to correlate with performance on higher

order cognitive activities that were assumed to depend on working memory. The issue of how to assess working memory has been addressed in different ways. One approach that has gained popularity among European researchers is to examine dual-task performance within the context of Baddeley's (1986) multi-component model of working memory. Another predominantly North American approach, has been to

examine individual differences in working memory capacity using a variety of comple span tasks designed to resemble the working memory demands of complex cognitive

activities. The third, most recent approach is to examine the factor structure unde

performance on a variety of working memory tasks and the predictive value of latent

constructs for higher-level cognitive performance using structural equation modelin These approaches are discussed in more detail below.

1.5.1 Dual-task performance

Since the development of Baddeley and Hitch's (1974) multi-component model of working memory, the dual-task paradigm has been widely used to specify the individual subcomponents of working memory implicated in the performance of

complex cognitive tasks. In the dual-task procedure, a cognitive task of interest i performed concurrently with a secondary task that is assumed to tap one ofthe 19

subcomponents of working memory. The assumption is that if performance ofthe

secondary task disrupts performance on the primary task compared to when each task is performed alone or with a different secondary task, then the subcomponent tapped by

the secondary task must be involved in the performance ofthe cognitive task of inter Baddeley and his colleagues have been successful in using this methodology to

demonstrate the differential involvement of each subcomponent in a range of cognitive

tasks. For example, articulatory suppression has been shown to interfere with a numbe

ofthe hallmark effects thought to reflect the operation ofthe phonological loop, suc the phonemic similarity and word length effects (Baddeley, 1986; Baddeley, Lewis, & Vallar, 1984; Richardson, Greaves, & Smith, 1980). As a result, articulatory

suppression has become the method of choice for disrupting the articulatory component ofthe working memory model. Tasks assumed to tap the visuospatial component of

working memory typically require the continuous tapping or visualisation of a spatial sequence (Logie, 1995; Logie, 1996). These tasks have been shown to selectively impair performance on visuospatial tasks, such as memory for visual matrix patterns (Logie et al., 1990) and the mental rotation of visual objects (Logie, 1995). The identification of a task that exclusively taps the central executive has proven to be more difficult. However, the random generation task, which requires the

generation of a random sequence of numbers or letters, has recently been adopted as a central executive task because successful performance requires the continuous monitoring of potential responses in order to prevent the generation of habitual sequences (Baddeley, 1996). Support for the selective involvement ofthe central executive in the random generation task comes from a study by Gilhooly, Logie, Wetherick and Wynn (1993) who demonstrated that a syllogistic reasoning task, assumed to place heavy demands on the central executive, was disrupted by a 20

concurrent random generation task but not by articulatory suppression or spatial tapping. The dual-task paradigm has been useful in determining the role that each

subcomponent plays in the performance of a range of complex activities, such as ver fluency (Baddeley, 1996), playing a computer game (Logie, Baddeley, Mane, &

Donchin, 1989), mental arithmetic (Logie, Gilhooly, & Wynn, 1994), and playing ches

(Baddeley, 1993). However, Hegarty, Shah, and Miyake (2000) have recently suggested that the logic ofthe dual-task procedure may not hold when applied to the central

executive. They found that secondary executive tasks produced the largest performan decrement on the task considered to be the least demanding ofthe central executive smallest decrement on the most demanding executive task, which is opposite to the pattern of results predicted by dual-task logic (Hegarty et al., 2000). Hegarty et (2000) argued that the pattern of results suggested the involvement of two related

factors that limit the application ofthe dual-task procedure, namely a response sel bottleneck and strategic trade-offs between primary and secondary tasks, and recommended that these constraints be considered when applying the dual-task procedure to the examination of central executive involvement. However, Hegarty et (2000) confirmed that if used appropriately, the dual-task could still be a useful informative paradigm for examining the role of working memory in cognition.

1.5.2 Complex span tasks

Over the last two decades of working memory research, the study of individual differences has become a popular method of determining the role of working memory capacity in the performance of various complex cognitive activities. A variety of complex span tasks have been developed as measures of an individual's working memory capacity (Case et al, 1982; Daneman & Carpenter, 1980; Shah & Miyake, 21

1996; Turner & Engle, 1989). These tasks incorporate both storage and processing and

are believed to reflect the limited capacity ofthe working memory system available t the individual. One ofthe original complex span tasks was a sentence span task developed by Daneman and Carpenter (1980), in which participants were required to

read a series of sentences aloud and then recall the last word of each sentence in t

series in correct serial order. The number of sentences presented was increased until

participant could no longer recall all the words in a series correctly. An individua

reading span was taken to be the longest series of sentences for which all the final could be correctly recalled. Daneman and Carpenter (1980) demonstrated that this

reading span measure correlated significantly with a number of reading comprehension measures, whereas a traditional word span task did not. A meta-analysis performed by Daneman and Merikle (1996) supported this finding; namely that complex span tasks which tap a combination of storage and processing are better predictors of language abilities than tasks that tap storage only. Although it has been argued that the reading span measure reflects a languagespecific system (Daneman & Tardif, 1987), Turner and Engle (1989) developed a

variation ofthe sentence span task that involved simple mathematical operations as t processing component and demonstrated that this span task correlated with reading comprehension as well as the reading span measure. Turner and Engle (1989) concluded

that the nature ofthe processing component was not crucial to the predictive power o the complex span task and argued that the span measure reflected a general working memory capacity. This conclusion is supported by the well-reported finding that

individual differences in working memory capacity as measured by these tasks reliabl predicts performance on a range of higher-order cognitive tasks including reasoning (Kyllonen & Christal, 1990), language comprehension (Just & Carpenter, 1992), 22

complex learning (Kyllonen & Stephens, 1990), strategic m e m o r y retrieval (Rosen &

Engle, 1997), following directions (Engle, Carullo, & Collins, 1991), and notetaking (Kiewra & Benton, 1988). Despite the popularity ofthe complex span tasks, Waters and Caplan (1996) have criticised the typical working memory span procedure of measuring recall performance alone without taking account of processing efficiency. They argued that

sentence span tasks are unreliable unless both the processing and recall components measured and taken into consideration (Waters & Caplan, 1996). However, Conway and

Engle (1996) found that equating the difficulty of processing on a complex span tas not affect the relationship between the working memory span measure and reading comprehension. Similarly, Engle, Cantor, and Carullo (1992) demonstrated that the correlations between performance on various complex span tasks and a measure of reading comprehension were not reduced when the time spent on the processing

component of each span task was statistically controlled for. These findings sugges processing efficiency is not the crucial determinant ofthe predictive ability of a complex span task. Conway and Engle (1996) argued that switching attention between

storage and processing operations was the critical determinant of working memory spa

performance, as it required the individual to engage in controlled effortful process They concluded that complex span performance reflected the capacity for controlled processing and that individual differences in working memory capacity would only be evident in tasks that demand controlled attention (Conway & Engle, 1996). Thus, complex span tasks have proved to be an invaluable tool for examining the role of working memory capacity in complex cognition, however, the specific mechanism underlying the predictive ability of these tasks is still open to question.

23

1.5.3 Structural equation modeling

One exciting new approach to examining the functions of working memory that

is gaining popularity among cognitive researchers is structural equation modeling. This technique is used to statistically extract the shared variance among selected tasks to

form a latent variable that represents the underlying construct common to the tasks. Th

relationship between the resulting latent variable and other similarly extracted latent

constructs or the contribution ofthe latent variable to performance on different comple cognitive tasks can then be examined. This latent variable approach has a number of

advantages over the typical individual differences approach in that latent variables ar thought to reflect more pure measures ofthe target function or construct than manifest variables, and so, should avoid the problem of task-specific idiosyncrasies (Miyake et al., 2000). Moreover, the multidetermined nature of most complex 'executive' type

tasks means the interpretation of results from correlational and factor-analytic studie

often difficult and arbitrary. Structural equation modeling can alleviate these difficu

by examining the unique contribution of each latent variable to complex cognitive tasks or constructs. Engle et al. (1999b) used this approach to examine the relationship between latent variables representing working memory, short-term memory and general fluid intelligence. They found that working memory and short-term memory were highly related but distinguishable constructs and that working memory was closely related general fluid intelligence independently of short-term memory (Engle et al, 1999b).

Miyake et al. (2000) also used structural equation modeling to examine the relationship between three executive functions and the contribution of these functions to a set of

frequently used executive tasks. They demonstrated that the three executive functions o

mental set shifting, information updating and the inhibition of prepotent responses wer 24

separable but related functions that shared some c o m m o n underlying mechanism. The

three executive functions were also shown to contribute differentially to performanc complex executive tasks, thus providing insight as to what the executive tasks commonly used to assess "executive processing" actually measure (Miyake et al.,

2000). This approach has also been used to examine both the structural and functional components underlying measures of working memory capacity (Oberauer, SuB, Schulze, Wilhelm, & Wittmann, 2000). Structural equation modeling is a flexible approach to examining relationships among variables and can often be used to

investigate research questions that are difficult to address with conventional stati techniques.

1.6 Summary Working memory is currently conceptualised as an integrated system responsible for the temporary maintenance and active manipulation of information

relevant for the task at hand. In contrast to the more traditional models of a shortstore, working memory models place greater emphasis on attentional mechanisms

responsible for the control and regulation ofthe system as a whole. Working memory is

assumed to be limited in capacity, however, the nature ofthe limitations is still op debate. A number of working memory theorists from a variety of approaches have suggested that the capacity of working memory is intimately linked to the efficiency

or capacity for inhibitory processing. Moreover, research into the control mechanisms of working memory has identified inhibition as one ofthe core executive processes. Thus, it seems that there is some evidence of a relationship between working memory

and the capacity for inhibition, however, the exact nature of this relationship and w

meant by the term 'inhibition' remains unclear. The following chapter will review the 25

constructs of inhibition and interference in relation to cognitive processing with a view to understanding how inhibition may be related to working memory performance.

26

Chapter 2 : Inhibition and Interference in Cognition 2.1 Development ofthe constructs of inhibition and interference

The constructs of interference and inhibition have their origins in classical interference theory, which was prominent in the late 19th century (Dempster, 1992). Interference theory was based on the notion that cognitive elements were connected

associative bonds and these associations could compete with one another and inhibit suppress the activation of competing elements (Dempster, 1992). Interference theory gained popularity with the development ofthe Brown-Peterson paradigm (Brown, 1958;

Peterson & Peterson, 1959) and the subsequent proliferation of studies demonstratin the impact of proactive interference on forgetting (Bennett, 1975; Murdock, 1961;

Wickens, 1970). Following this early verbal learning research, the term interference

used to refer to the direct cause ofthe observed decrement in performance due to ta

irrelevant material (Dempster, 1995), and inhibition was thought of as the mechanism by which interference was resolved (Arbuthnott, 1995). However, there was little

agreement as to the nature of these mechanisms and their explanatory value was call into question. As a result, the constructs of inhibition and interference received

attention from researchers for a number of decades until a recent resurgence in int inspired by converging lines of research into cognitive development and aging, neuropsychology, and individual differences in working memory ability.

2.1.1 Cognitive development and aging

A number of contemporary models of development and aging have incorporated

inhibitory mechanisms to account for the developmental changes in children's cognit processes and the cognitive decline observed in later life. Many of these theories 27

postulate a central pool of limited mental resources that is available for the execution of various cognitive operations and storage of information (Bjorklund & Harnishfeger, 1987; Case et al., 1982). According to these models, the resources available to an individual remain constant with age and developmental improvements in cognitive performance are attributed to increases in processing efficiency, which frees up resources for additional storage and/or the execution of other cognitive processes (Harnishfeger, 1995; Harnishfeger & Pope, 1996). Bjorklund and Harnishfeger (1990)

extended the limited resource model to include inhibitory processes responsible for the

maintenance of efficient cognitive processing through the suppression of task irreleva

information. In this model, inhibition is described as an active process that restrict spread of activation to the irrelevant material (Harnishfeger, 1995). Bjorklund and Harnishfeger (1990) argued that inhibitory processes become more efficient with age and that young children's immature inhibitory mechanisms allow irrelevant information to enter and be maintained in their limited working memory space resulting in less

efficient cognitive processing. In support of their argument, Harnishfeger and Bjorklu (1993) reported a series of memory experiments in which younger children produced

more task irrelevant intrusion errors than older children. They also provided convergi evidence from a variety of paradigms that inhibitory efficiency improves with age and argued that inhibition might be an explanatory mechanism for general cognitive development (Harnishfeger & Bjorklund, 1993). A similar argument has been proposed by researchers interested in cognitive aging to explain the cognitive decline often observed at the other end ofthe lifespan continuum. Hasher and Zacks (1988) proposed a model of cognitive aging that also featured inhibition as a means of explaining the poor performance of older adults on a range of cognitive tasks. They believed that inhibition was a central mechanism responsible for 28

controlling the contents of working m e m o r y by limiting the information that gains entry into working memory and suppressing information that is no longer relevant to the current task demands (Hasher & Zacks, 1988). Hasher and Zacks (1988) argued that older adults have less efficient inhibitory processes, which makes them more likely to activate and maintain irrelevant information in working memory. Consistent with this framework, Hartman and Hasher (1991) found that older adults were more likely than younger adults to maintain both the target words and the disconfirmed words from the ends of garden path sentences, even though the disconfirmed endings were irrelevant to the task. In another study, Zacks, Radvansky and Hasher (1996) examined the performance of older adults on a directed forgetting task in which participants were presented with lists of words and were instructed to either remember or forget each word following its presentation. They found that compared to younger adults, older

adults recalled and recognised more items that they were instructed to forget. Zacks et al. (1996) suggested that this was a result of deficient inhibitory mechanisms in the elderly, which would normally serve to prevent further processing and maintenance of the to-be-forgotten items. Further evidence of reduced cognitive inhibition in the elderly has been demonstrated using the negative priming paradigm. In this paradigm, stimuli that are presented as distractors on a prime trial are then presented as target stimuli on the subsequent probe trial. This typically results in increased response times relative to conditions in which there is no relationship between items on successive trials. This increase in response time has been termed the 'negative priming effect' (Tipper & Cranston, 1985). The negative priming effect has been widely used as a measure of

cognitive inhibition, however, the extent to which this measure reflects inhibition is issue under current debate in the literature (for further detail see 2.4.1 Negative 29

priming). Using this paradigm, Hasher, Stoltzfus, Zacks and R y p m a (1991) found that older adults failed to show negative priming, which they argued was evidence ofthe reduced efficiency of cognitive inhibition in the elderly. Hasher and Zacks (1988)

argued that this inefficient inhibitory processing was the main factor contributing to impaired performance ofthe elderly on various tasks reliant on working memory. Thus, according to the models of cognitive development and aging reviewed here, efficient inhibition is necessary for efficient working memory processing.

2.1.2 Neuropsychology

The frontal lobes are the most recent structures to appear in the evolution ofthe

brain and as a result, they have historically been associated with the ability to perf

complex executive functions, including the capacity for inhibition (Fuster, 1991; Luri 1966). However, recent advances in our understanding ofthe human brain have strengthened the evidentiary link between the operation ofthe frontal lobes and inhibitory processing. Dempster (1992; 1995) presents converging evidence in support

of this association. First, a similar pattern of performance deficits is found with yo

children, older adults and frontal lobe patients on a range of executive tasks that re the suppression of irrelevant stimuli (see Dempster, 1992, for a review). Second, the frontal lobes are the last region ofthe brain to develop with changes in myelination,

cortical fissuration and synaptic density continuing into adolescence. The frontal lob are also the first to show signs of involution during late adulthood (see Dempster,

1992). These changes in brain structure correspond closely to the changes in inhibitory function described in the cognitive development and aging literature, however, this of course may be only incidental.

30

Advances in neuroimaging techniques have also contributed to the reinvigoration of research into the role of inhibition and interference in complex

cognitive activities. A number of recent neuroimaging studies have been concerned with isolating the specific components of working memory and identifying the brain regions

associated with those components (Awh et al., 1996; D'Esposito et al., 1995; Jonides e al., 1998a; Jonides, Smith, Koeppe, & Awh, 1993; Smith & Jonides, 1998). Using positron emission tomography (PET), Awh and colleagues (1996) provided evidence of

a dissociation between the brain regions activated by storage and rehearsal processes a verbal working memory task. In the verbal memory task, participants were presented

with a continuous sequence of letters and were required to indicate whether each lette presented was the same as the letter that was presented two letters previously in the sequence. In addition, participants completed two control tasks. In one of these, the search control task, participants were again presented with a continuous sequence of

letters and were required to decide if the letter presented was the same as the first presented in the sequence. This task involved the same perceptual and response requirements as the verbal memory task, but the memory load was substantially reduced. In the second control task, the rehearsal control, participants were again presented with a continuous sequence of letters, but were simply required to indicate

when each letter appeared and repeat the letter silently until the next letter appear This task involved similar rehearsal and response requirements as the verbal memory task. By subtracting the activations associated with the search control task from the verbal memory task, Awh et al. (1996) were able to isolate the brain regions involved the storage and rehearsal operations of working memory. These included anterior brain

regions associated with speech planning and execution, most notably, Broca's area, the premotor cortex and the supplementary motor area, as well as regions in the posterior 31

parietal cortex. B y subtracting the activations associated with the rehearsal control task

from the verbal memory task, Awh et al. (1996) found a significant loss of activation i Broca's area and the premotor cortex, whereas the activation in the posterior parietal cortex remained significant. Based on these findings, Awh et al. (1996) argued that Broca's area and the premotor cortex were involved in subvocal rehearsal processes, whereas the posterior parietal cortex was involved in phonological storage. Using a similar technique, Jonides, Smith, Marshuetz, Koeppe, and Reuter-Lorenz (1998b) examined the brain activation associated with inhibitory processing in verbal working memory. Participants were presented with an item-recognition task in which some of

the trials required the inhibition of a prepotent but incorrect response that had rece been active in memory. Jonides et al. (1998b) found that responses on these trials

produced reliable activation in the left lateral prefrontal cortex (PFC) compared to t that did not involve inhibition. One limitation of PET is that the activations measured reflect all events that

occur during a trial. That is, it lacks the temporal resolution to measure activations response to specific events that occur at specific time points within a trial (Smith & Jonides, 1998). However, recently developed techniques in functional magnetic resonance imaging (fMRI) allow images to be analysed on a single trial basis which can be further isolated into different temporal periods within a trial (Smith & Jonides, 1998). Using this technique, D'Esposito, Postle, Jonides and Smith (1999) conducted a similar study and analysed activation during different temporal components ofthe task. They found increased activation in the left ventrolateral PFC, which only became evident with the onset ofthe probe item for recognition. D'Esposito et al, (1999) proposed the existence of a PFC-mediated mechanism responsible for the resolution of interference caused by competition from previous target sets. Furthermore, they 32

concluded that these interference resolution processes were dissociable in time from the processes involved in the encoding and maintenance of information (D'Esposito et al., 1999). Developments in neuroimaging techniques have resulted in exciting advances in our understanding ofthe cognitive architecture of working memory. Having said that, neuroimaging techniques are limited in that they can only provide an indirect measure of neural activity by detecting changes in blood flow to particular areas ofthe brain. However, the obvious benefit of this type of research is the potential to isolate the contribution of specific processes, such as inhibition, to performance on a working memory task.

2.1.3 Individual differences in working memory and general cognitive ability

A number of recent studies examining individual differences in working memory have also emphasised the role of inhibitory processes. As reviewed in the previous chapter (1.4.1 Capacity limitations), Conway and Engle (1994) argued that individual differences in working memory capacity reflected differences in the

attentional resources necessary for the inhibition of task-irrelevant information. In a further study, Engle, Conway, Tuholski and Shisler (1995) examined this proposal more

directly using a negative priming paradigm similar to that discussed previously (see a 2.4.1 Negative priming) as a measure of cognitive inhibition. They argued that if

inhibition required attentional resources, then drawing on those resources with another attention-demanding task would reduce an individual's ability to inhibit irrelevant stimuli, which would be evidenced by a reduced negative priming effect. Participants

were presented with red and green letter pairs and were required to name the red target

letter and ignore the green distractor letter. On the negative priming trials, the ign distractor letter was presented as the target letter on the subsequent display, which 33

typically leads to the slower naming latencies k n o w n as the negative priming effect.

These trials were interspersed with words that were to be remembered for recall at th end of a trial sequence. Engle et al. (1995) predicted that as memory load increased

towards the end of a trial, the negative priming effect would be reduced. As expected

large negative priming effect was evident before the presentation ofthe first word t remembered and decreased as the number of words in memory increased, eventually becoming a facilitatory effect at the end ofthe trial. Engle et al. (1995) concluded inhibition is resource demanding and when the required resources are not available, inhibition is no longer possible. Conway, Tuholski, Shisler and Engle (1999) extended this study to examine the performance of participants classified as high or low in working memory capacity on the negative priming task described above with both a verbal and nonverbal memory

load interspersed between the trials. In both the verbal and nonverbal experiments, a significant negative priming effect was evident before the presentation ofthe first memory item when the memory load was zero, however this effect was eliminated with the introduction of a memory load (Conway et al, 1999). This was taken as evidence

that the processes involved in negative priming are dependent on a domain-free pool o

resources. Furthermore, analysis of individual differences in working memory capacity revealed that participants classified as high in working memory capacity showed

significant negative priming, whereas participants classified as low in working memo

capacity did not. This result provides further support for the suggestion that indiv differences in working memory capacity correspond to the ability to manage task irrelevant information efficiently (Conway et al., 1999). Taken together, the studies Engle and his colleagues have highlighted the intimate relationship between working memory and the attentional resources necessary to overcome interference. 34

Another theory that has been applied to the study of individual differences in

cognitive ability is the resistance to interference theory, which arose as an alternat

basic activation-resource and strategic explanations of cognitive development and agin

(Dempster, 1992). Resistance to interference refers to the ability to inhibit irrelevan

information and is assumed to be a primitive feature ofthe cognitive system that canno be explained in terms of other cognitive processes (Dempster & Corkill, 1999). Another assumption of this framework is that individual differences in the ability to resist interference depend upon the efficiency ofthe frontal lobes ofthe brain (Dempster & Corkill, 1999). As a result, patients with frontal lobe lesions often show impaired performance on tasks that are susceptible to interference such as tasks that involve competition among stimuli or responses, distractor activity, secondary tasks, multiple

trials, or trials in which previously relevant stimuli are no longer relevant (Dempster 1992; Dempster & Corkill, 1999). These include the Wisconsin Card Sorting Task (WCST), the Brown-Peterson task, the Stroop task, and the A-B, A-Br task. In the WCST, individuals must categorise cards according to a sorting rule determined by the experimenter and avoid categorisation based on previous sorting rules. The BrownPeterson task requires participants to maintain a list of items in memory whilst performing an interpolated distractor activity. A series of these memory trials are usually presented with similar stimuli in each trial to induce proactive interference across trials. In a related task, the A-B, A-Br task, stimulus and response terms from

initial learning list are recombined to form new pairs on successive lists. Interferen typically measured in terms ofthe decline in performance across successive lists. However, the most popular measure of interference sensitivity would undoubtedly be the Stroop task. In this task, competition from the word-reading response must be overcome in order to name the colour ofthe ink in which the word is presented. 35

Dempster and Corkill (1999) report a factor-analytic study designed to examine the relationship between three of these interference tasks (WCST, Stroop, and A-B, ABr) and two measures of general mental ability (Raven's matrices and the Block Design Subtest). The principal components analysis produced three resultant factors, one that consisted ofthe WCST variables, one that consisted ofthe paired-associate variables,

and one that consisted ofthe Stroop variables, intrusion errors on the paired associate task, the Raven's variables, and the Block Design score. Dempster and Corkill (1999) argued that the results of this study were consistent with the idea that individual

differences in resistance to interference, as measured by the Stroop test and intrusion

errors, are an important factor in general cognitive ability. However, having said that

Dempster and Corkill (1999) found little other evidence in the literature in support o specific association between interference and general mental ability. They suggested that a more detailed framework that incorporated task variables and specific processes

into the analysis of individual differences in susceptibility to interference and cogn ability was required (Dempster & Corkill, 1999).

2.2 Classification of Inhibition and Interference Processes

As Harnishfeger (1995) pointed out, it is not clear whether there is a single

inhibitory process that applies to all the paradigms that make use ofthe term, or if t are a number of separable processes that may or may not share a common underlying mechanism. One approach to delineating the boundaries ofthe inhibition construct is to propose separate processes that have different operating characteristics. For example, inhibitory processes can be distinguished according to the different psychological constructs they exert control over (cognitive or behavioural), or the level of effort required to engage the inhibitory processes (automatic or intentional). 36

Behavioural inhibition refers to the control of overt behaviour and is thought to be the mechanism responsible for delaying gratification, impulse control and motor

inhibition (Harnishfeger, 1995). The ability to delay responding can be assessed throug various differential reinforcement paradigms such as those of Mischel and colleagues (Mischel, Ebbesen, & Raskoff-Zeiss, 1972; Mischel, Shoda, & Peake, 1988). Motor inhibition refers to the ability to control or interrupt an ongoing motor response and often examined with the stop-signal paradigm (Logan, 1994). In this paradigm, participants perform a speeded cognitive task that requires an overt response, such as

lexical decision or a forced-choice discrimination task. On a certain percentage of tri participants are given a stop-signal, which indicates that they should withhold their

response on that trial. Based on the interval at which participants are unable to withh their response, the time required to process a stop-signal has been estimated to be approximately 200ms, and is not affected by the nature ofthe primary task (Logan, 1994). This suggests that a general inhibitory mechanism may underlie performance on this task. In addition, children with attention deficit hyperactivity disorder (ADHD), who are known to have difficulties with impulse control, have demonstrated impaired performance on this task suggesting that it does tap one ofthe mechanisms responsible for the control of behaviour. As the term implies, cognitive inhibition refers to the control of cognitive

contents or processes (Harnishfeger, 1995). At a more specific level, this construct can

be further distinguished in terms of whether the inhibition is automatic or intentional Intentional inhibition refers to a deliberate suppression process invoked to exclude irrelevant information from consciousness (Harnishfeger, 1995; Harnishfeger &

Bjorklund, 1993). The best example of a task used to assess intentional inhibition is t directed-forgetting paradigm. In this paradigm, naive participants are instructed to 37

forget a set of previously learned target items, but are then given a recognition or recall

test for all items. The efficiency of inhibition is measured in terms ofthe number of to be-forgotten items that are recalled or recognised relative to to-be-remembered items. Although current theories postulate that multiple mechanisms are responsible for the

directed forgetting effect, the inhibition of retrieval of to-be-forgotten items is ess to prevent them from interfering with the rehearsal and retrieval of to-be-remembered items (Zacks & Hasher, 1994). In contrast, automatic or unintentional inhibition has been defined as the preconscious suppression of irrelevant information that becomes activated alongside relevant information (Harnishfeger, 1995; Wilson & Kipp, 1998). Automatic inhibition has typically been assessed using lexical-ambiguity and/or negative priming tasks. In

these paradigms, the suppression of irrelevant distractors or inappropriate meanings of ambiguous words occurs automatically without conscious awareness or intention. For example, Simpson and Kang (1994) found that naming times for probe targets that were associated with a semantically related prime were slowed if an alternative meaning of that prime was cued on the previous trial, suggesting that the meaning associated with semantically related prime had been suppressed. From a different line of research, the negative priming paradigm has been a popular measure ofthe inhibitory processes

involved in selective attention tasks. In this paradigm, participants select and respon

a target stimulus in the presence of one or more distractors. Negative priming refers to

the typical finding that responses to target stimuli are slower if the target was prese

as a distractor item on the previous trial (see 2.4.1 Negative priming, for more detail) Tipper and his colleagues have argued that the increased response times to the target reflect the process of overcoming the inhibition that was applied to the item on the previous trial (Tipper, 1985; Tipper & Cranston, 1985; Tipper & Driver, 1988). 38

Although the lexical-ambiguity and negative priming paradigms are both assumed to

involve the automatic suppression of irrelevant items, it is yet to be determined whet the same inhibitory process operates in both. Two mechanisms by which automatic inhibition may be achieved are lateral inhibition of units within the same layer of an

associative network and self-inhibition of a unit back to baseline immediately followi

its activation (Arbuthnott, 1995). These mechanisms will be discussed in more detail in the section assigned to inhibitory mechanisms below (see 2.5 Mechanisms of inhibition). A number of authors have also emphasised the distinction between cognitive inhibition and interference (Harnishfeger, 1995; Nigg, 2000; Wilson & Kipp, 1998). This distinction has, in part, been derived from Dempster's resistance to interference theory (Dempster, 1991; Dempster, 1993; Dempster & Corkill, 1999). Based on this theory, Harnishfeger and colleagues (Harnishfeger, 1995; Wilson & Kipp, 1998) refer to interference as performance decrements caused by multiple sources of competing information. In a more recent taxonomy of inhibitory processes, Nigg (2000) refers to

interference control as the ability to prevent interference from competing, distractin interfering stimuli. Both theorists argue that interference control is implicated in

selective attention tasks such as the Stroop task and flanker tasks (Harnishfeger, 1995

Nigg, 2000). In these tasks, the participant must select the task-relevant stimuli in t face of distraction from irrelevant stimuli that are competing for response. However, what is not clear is the nature ofthe mechanisms involved in resolving this response

competition. Harnishfeger (1995) argues that interference does not necessarily involve the active suppression of competing stimuli, but may result from bottlenecks, during

which selective procedures must isolate the response that will be produced. In contrast

Nigg (2000) refers to the "effortful interference control" involved in the Stroop task 39

the intentional inhibition of a competing automatic motor/vocal response (p. 223). Furthermore, Dempster himself describes resistance to interference as "the ability to ignore or inhibit irrelevant information" (Dempster & Corkill, 1999, p.397). Thus, the

distinction between cognitive inhibition and interference often becomes a little blurred It seems that most theorists either implicitly or explicitly incorporate some type of inhibitory mechanism to explain the control of interference produced by response competition. The main question that remains open to debate is whether this inhibitory process is automatic (Wilson & Kipp, 1998), or involves the effortful suppression of

interfering stimuli (Nigg, 2000). This issue will be returned to in the following section

2.3 The relationship between interference and inhibition

It is evident that the constructs of inhibition and interference have become increasingly popular in a number of areas of psychological research, however, some researchers refer to inhibitory processes while others pose their theories in terms of susceptibility to interference. As a result, both terms have often been used interchangeably in the literature, which has led to confusion over the underlying mechanisms involved and the relationship between interference and inhibition. This

stems from the lack of a clear theoretical definition and empirical operationalisation o the constructs. As discussed in the previous section, the term interference typically

refers to performance decrements caused by irrelevant information or distracting stimuli (Dempster & Corkill, 1999), whereas inhibition is used to refer to an active suppression process (Harnishfeger, 1995; Harnishfeger & Bjorklund, 1993). Confusion arises because a number ofthe paradigms originally used to examine interference have been modified to investigate inhibition (e.g. Stroop task) (Harnishfeger, 1995). This also

raises questions about the nature ofthe relationship between inhibition and interference 40

and whether they are separable processes that can be manipulated independently, or whether they are interdependent with a change in one resulting in a corresponding change in the other. Evidence for the interdependency of inhibition and interference comes from studies that demonstrate an inverse relationship between interference sensitivity and negative priming. This has most often been demonstrated in comparisons of groups who are more or less susceptible to interference. For example, McDowd and Oseas-Kreger (1991) found that older adults showed increased interference and diminished negative

priming relative to younger adults on a standard negative priming task. Similarly, usi a Stroop paradigm, Tipper, Bourque, Anderson and Brehaut (1989) found significantly

larger interference effects in children relative to adults, but significantly larger n priming effects in adults relative to the children who failed to show any negative priming. A similar pattern has also been demonstrated by patients with schizophrenia (Beech, Baylis, Smithson, & Claridge, 1989a; Beech, Powell, McWilliam, & Claridge, 1989b). These findings suggest that the mechanism responsible for negative priming is potentially the same mechanism by which interference is resolved (Houghton & Tipper, 1994). However, as Neill, Valdes and Terry (1995) point out, exactly how inhibition

relates to interference within an individual is uncertain. According to Neill et al. (1 inhibition may be reactive to the degree of interference experienced, or anticipatory that it occurs independently ofthe interference actually encountered. For example, an

irrelevant distractor that causes minimal interference would require minimal inhibitio

and so, if inhibition is reactive, then negative priming should be directly associated

degree of interference (i.e. less interference, less negative priming) (Neill & Valdes, 1996; Neill et al, 1995). Alternatively, if inhibition is anticipatory and operates 41

independently of interference, then a constant level of inhibition applied to a weak distractor would produce more negative priming than if applied to a strong distractor. Furthermore, if the activation of a strong distractor outweighs the level of inhibition applied, then positive priming may be evident. Thus, according to Neill et al. (1995),

negative priming in this situation should be inversely related to interference (i.e. les

interference, increased negative priming). In contrast, Fox (1994) argued that an inverse relationship between interference and negative priming would be indicative of a causal

relationship whereby an increase in the inhibition applied to the distractors would resu in less interference but increased negative priming. According to Fox (1994), an independence of interference and inhibitory processes would be evidenced by an inconsistent relationship that is neither positive nor negative. The findings in the literature relevant to this issue have been mixed. Neill (1995) reports a series of unpublished studies in which he and his colleagues have generally found that variables that increase interference also increase negative priming. For example, Neill and Lissner (1988, cited in Neill & Terry, 1995) found that both negative priming and interference in a letter-matching task increased when the proportion of trials with incompatible distractors was reduced. They argued that if inhibition were anticipatory, then negative priming should have been strongest when the proportion of

incongruent trials was high and participants were expecting interference. In contrast, t

pattern of results provides support for the reactive nature of inhibition. Further evide comes from a study by Fox (1994, Experiment 1), in which reducing the spatial separation between targets and distractors produced an increase in both interference and negative priming. However, Fox (1994) argued that the two effects showed different

patterns of change across the different spatial separations, suggesting that they were i fact independent processes. This claim was further supported in a separate study in 42

which Fox (1995b, Experiment 3) demonstrated that pre-cueing a target location significantly reduced interference from incompatible distractors, but showed no influence on negative priming effects. The independence of interference and inhibition has also been demonstrated by Driver and Tipper (1989), who found that non-

interfering distractors produced comparable negative priming to interfering distractors Thus, there appears to be some evidence for the independence of interference and negative priming effects within individuals. Although this apparent dissociation between interference and inhibition at the empirical level may be taken as evidence for the involvement of separate underlying mechanisms, this is not necessarily the case. If both excitatory and inhibitory

mechanisms are involved, then it may be that certain contextual manipulations within an experiment result in changes in the excitatory component but have no effect on the inhibitory component (Fox, 1995a). In line with this proposal, a neural network model of selective attention developed by Houghton and Tipper (1994, see 2.5 Mechanisms of

inhibition below), demonstrates that even though inhibition is a mechanism of selection that determines interference from irrelevant distractors, dissociations between interference and inhibition as measured by negative priming are also possible. The advantage of such models is that they allow more precise predictions about the relationship between inhibition and interference to be made. As discussed by Fox (1994), another explanation for the inconsistent relationship between interference and inhibition is that the implementation of inhibition takes time. Inhibition applied to

distractors may lead to a reduction of interference and the presence of negative primin on subsequent trials involving those distractors, but may not be implemented fast enough to be effective in the trial that it is applied. Thus, certain experimental manipulations such as pre-cueing the prime target may influence the speed with which 43

inhibition can be implemented, and consequently, the amount of interference observed. The relationship between inhibition and interference would then be expected to be highly context dependent and to vary according to the demands ofthe task. This notion is consistent with the finding that inhibition is determined to a large extent by the behavioural goals ofthe task (Tipper, Weaver, & Houghton, 1994), and strengthens the conception of inhibition as a flexible process that adapts according to current task demands. An alternative explanation for some ofthe inconsistencies in the negative priming literature is provided by Lavie's (1995) "perceptual load" model of selective attention. According to this model, perceptual processing is limited in capacity, but proceeds automatically until the system reaches this limit. Once the capacity limit is exceeded, selection ofthe relevant information to be processed will be required. When the relevant stimuli do not demand all ofthe available attentional capacity, any spare capacity is automatically allocated to the irrelevant stimuli. Lavie (1995) argues that allocation of attention to irrelevant stimuli cannot be actively prevented by inhibitory processes, and that these processes are more likely to be involved in post-perceptual operations such as memory, response selection and execution. Hence, in this view selection is a natural consequence of exceeding the capacity ofthe perceptual system. Early selection is predicted under conditions of high perceptual load that demand all available attentional capacity, and late selection is expected in situations of low perceptual load that leave spare capacity for the processing of irrelevant information

(Lavie & Fox, 2000). This leads to the counterintuitive prediction that interference from irrelevant distractors will be greater in situations of low rather than high perceptual

loads, because there will be more distractor processing under conditions of low load that do not exhaust attentional capacity. 44

Lavie (1995) provided support for the perceptual load theory of selective attention in a series of experiments that directly varied the perceptual load ofthe relevant processing and examined the interference effects from irrelevant distractors. Experiment 1, Lavie (1995) showed that increasing the number of items that had to be searched to identify the target resulted in less interference from an incompatible distractor that was clearly distinct from the target search set. However, this manipulation involved a noticeable change in the appearance ofthe display, which may have influenced the perceptual saliency ofthe distractor. In Experiment 2 A, Lavie (1995) avoided this problem by manipulating the processing demands involved in the

task whilst maintaining identical display characteristics. In this experiment, particip

made a choice response to a target letter in the presence of an irrelevant distractor th was either neutral or incompatible in relation to the target response. Perceptual load manipulated by way of a coloured shape presented next to the target, which determined whether the participant responded or withheld their response to the target. In the low load condition, participants responded or withheld their responses on the basis ofthe colour ofthe shape while in the high load condition, participants based their responses on the conjunction ofthe colour and type of shape presented. In accordance with the perceptual load hypothesis, the interference effect was significant only under the low load condition. In the final experiment, Lavie (1995) again demonstrated significant

interference under conditions of low perceptual load involving the detection of a circle

or line, the presence of which signalled that a response to the target was to be made. N interference was found in the high load condition that required the participants to identify the size and position ofthe circle or line to determine whether they were to

respond. Lavie's (1995) results provide strong evidence that the efficiency of selectiv attention is dependent on the perceptual load involved in the task. 45

Although Lavie (1995) argued that an active inhibition mechanism was not

necessary to explain selective processing under situations of high perceptual load, she acknowledged that such a mechanism may be required under situations of low perceptual load when spare capacity is automatically allocated to the processing of irrelevant distractors. The resulting prediction is that greater distractor inhibition required in situations of low perceptual load in which more distractor processing is encountered. Thus, consistent with a reactive view of negative priming, Lavie and Fox (2000) made the prediction that situations of low perceptual load would require active

inhibition mechanisms and thus lead to negative priming. In contrast, situations of high

perceptual load were expected to lead to reduced perception of irrelevant distractors a consequently, less negative priming. Lavie and Fox (2000) examined this prediction in another series of experiments in which they varied the perceptual load involved in processing the prime target and measured the resultant negative priming effect on the probe display. In the first three experiments, perceptual load was manipulated by

increasing the number of items in the target search set. As predicted, increases in the

perceptual load of target processing on the prime trial led to a significant decrease i negative priming on the probe display. Lavie and Fox (2000) argued that active inhibition may be important in avoiding response competition in situations of low perceptual load, however the need for this inhibition is reduced in situations of high

perceptual load. In these situations, selection occurs as a natural consequence of limi perceptual capacity (Lavie & Fox, 2000). Thus, perceptual load appears to play an

important role in visual selective attention and consideration of this factor may resol some ofthe apparent discrepancies between previous studies examining the relationship between interference and negative priming effects.

46

2.4 Measuring inhibition and interference

The preceding discussion identified a number of different methods for

measuring inhibition and interference. Deficient inhibition results in the activation maintenance and retrieval of irrelevant information. The measures typically used to assess inhibition include response times in negative priming paradigms, the activation

irrelevant information in lexical ambiguity tasks and intrusions in memory in directe

forgetting tasks. Interference reflects the extent to which response selection bottlene are produced by either concurrent task requirements in dual-task paradigms, or the

presence of distracting information in selective attention tasks that disrupts the sele of target information. In dual-task paradigms, interference is measured by examining performance decrements in dual-task compared to single task conditions, whereas in

selective attention tasks, interference is measured by a comparison of response times o both distractor-present and distractor-absent trials. Despite the number of methods available, one ofthe problems associated with the measurement of inhibition and interference is that the various tasks and paradigms were not developed from a psychometric standpoint, and therefore do not typically result in reliable scores or

produce normal distributions. However, this can be circumvented to a certain extent b

testing large samples of participants and including a sufficient number of experiment trials to produce stable results. The paradigm that has typically been viewed as the direct index of inhibitory processing and also provides a measure of distractor interference is the negative priming paradigm.

2.4.1 Negative Priming

Negative priming is demonstrated in the context of selective attention tasks in which the participant selects a target stimulus on the basis of some task-relevant cue 47

responds to that stimulus in the presence of one or more distractors. The critical manipulation is the relation between targets and distractors on consecutive trials. Negative priming is demonstrated on trials in which the target stimulus was the

distractor stimulus on the previous trial. The negative priming effect refers to the sl

reaction times and less accurate responding on negative priming trials relative to cont

trials in which there is no association between the targets and distractors on consecut trials. In some task contexts, these "control" trials provide a measure of distractor interference, which is inferred from longer response times on response-incompatible trials where the distractor is potentially task-relevant and therefore competes for response selection, compared to response-compatible trials where there is no response competition from distractors. Negative priming has been demonstrated with a range of

stimuli (pictures, words, letters, Stroop stimuli), and across a variety of task demands

(identity naming, lexical decision, spatial location), which suggests the generality oft effect (for reviews see Fox, 1995a; May, Kane, & Hasher, 1995). The dominant

explanation ofthe negative priming effect is that in the course of target selection, the activated representation ofthe distractor is suppressed or decoupled from response mechanisms (Dalrymple-Alford & Budayr, 1966; Neill, 1977; Neumann & DeSchepper,

1991; Tipper, 1985; Tipper & Cranston, 1985). If this distractor then becomes the target on the subsequent trial, the suppression applied to it on the previous trial must be overcome before a response can be made resulting in longer response times. Consequently, the negative priming effect is widely viewed as an index of inhibition. However, the underlying mechanisms thought to be responsible for negative priming has been the subject of continued debate. Although a complete review ofthe extensive negative priming literature is beyond the scope of this thesis (for reviews see Fox,

48

1995a; M a y et al., 1995; Neill & Valdes, 1996), the evidence relevant to the debate over the role of inhibition in negative priming will be examined. As discussed above, distractor inhibition is the most influential account ofthe

negative priming effect. According to the distractor inhibition theory, negative primi is produced by a dual-process selective attention mechanism, which involves

independent excitatory and inhibitory processes. The excitatory processes automaticall activate internal representations of target and distractor stimuli in parallel, while inhibitory processes actively suppress the internal representations of distractors or decouple them from response effectors (Neumann & DeSchepper, 1991; Tipper, 1985).

As described above, this inhibition is manifest in the delayed response to a distractor

item that subsequently becomes the target, that is, the negative priming effect. Tipper (1985) suggested that the negative priming effect may reflect a process of spreading inhibition, analogous to that of spreading activation. In Experiment 3, Tipper (1985) found significant negative priming effects using two superimposed line drawings of

common objects that were identified as a target or distractor on the basis of colour. T

interesting finding from this experiment was that when the target object (e.g. dog) was semantic associate ofthe previously ignored distractor object (e.g. cat), a similar negative priming effect was produced. Tipper (1985) suggested that the inhibition

applied to the internal representation of a distractor may spread to related concepts. effect was replicated by Tipper and Driver (1988) across symbolic domains (pictures and words), suggesting that negative priming operates at an abstract level of representation. Further support for the spreading inhibition hypothesis is provided in two studies by Neumann and his colleagues (Neumann, Cherau, Hood, & Steinnagel, 1993; Neumann & DeSchepper, 1992) that have demonstrated the existence of inhibitory fan49

effects using negative priming tasks and a Sternberg-type short-term m e m o r y scanning task. Neumann and DeSchepper (1992) demonstrated that the negative priming effect

was reduced as the number of distractors in the previous display increased. In additio a reversal of this effect to facilitation was produced when the task instruction emphasised speed over accuracy. Neumann and DeSchepper (1992) interpreted these

findings as reflecting a spreading inhibition mechanism that is limited in capacity an

operates on irrelevant information after this material has been automatically activate

Neumann et al. (1993) replicated this effect while avoiding the possible confound in h

earlier study of having the same number of distractors on both the prime and probe tri

They concluded that when the number of distractors increases, there is a dispersion of

inhibition resulting in less inhibition per distractor and consequently, less negative priming when one ofthe distractors becomes the subsequent target (Neumann et al., 1993). The results of Neumann et al. (1993) strongly implicate a spreading inhibition

explanation of negative priming. However, challenges to the distractor-inhibition theo comes from the work of Neill and Valdes (1992) and Neill, Valdes, Terry and Gorfein (1992), who demonstrated that negative priming was dependent on both the delay

between the prime and probe, and on the delay prior to the prime display. Neill et al. (1992) found negative priming was the greatest when the prime trial was temporally distinct from the previous processing episodes and the delay between the prime and

probe was short. In contrast, the least negative priming was found when the prime tria was not temporally distinct from the previous processing episodes and the delay between prime and probe was long. If negative priming is caused by the inhibition of distractors during target selection on the prime trial, it is not clear why the delay

the prime is critical. To account for these data, Neill and Valdes (1992) and Neill et 50

(1992) proposed the episodic-retrieval account of negative priming as an alternative to

the distractor inhibition view. According to episodic-retrieval theory, the presentatio a stimulus cues the retrieval ofthe most recent episode containing that stimulus. On prime and probe trials, distractors are encoded with "do not respond" tags, whereas

targets are encoded with "respond" tags. On negative priming trials, the target stimulu on the probe display cues the retrieval ofthe prior processing episode in which the stimulus was encoded with a "do not respond" tag, which conflicts with the current

response requirement (i.e. "respond"). As a result, a delay occurs while the conflict i resolved, which is evidenced as negative priming. Although many aspects of negative priming can be explained by episodic

retrieval (see Neill, 1997; Neill et al., 1992), there are a number of findings that ar

problematic for this view. Most notably, Conway (1999) failed to replicate the findings of Neill et al. (1992) in four experiments designed to systematically manipulate the interval prior to the prime display and also the interval between the prime and probe displays. Conway (1999) showed that negative priming was not affected by the delay

prior to the presentation ofthe prime display, or the interval between the prime displa

and probe display, and argued that there was little evidence from studies examining the

time-course of negative priming that uniquely supported the episodic retrieval theory o

negative priming. This finding is important because the time-course data of Neill et al (1992) provided the basis ofthe original episodic retrieval theory. In addition, the finding that under certain conditions, such as an emphasis on speed of responding over accuracy, ignored distractors produce facilitation on the subsequent trial is also problematic for episodic retrieval theory (Neumann & DeSchepper, 1992, Experiment

2). If, as episodic retrieval theory suggests, negative priming is caused by the retrie

of an item that has been previously been associated with an "ignore" tag, then there is 51

no reason that a change in instructional emphasis should influence this retrieval process

or subsequent conflict resolution. In contrast, an inhibitory account of negative primi can explain these findings in terms ofthe time period over which negative priming develops. Neill and Westberry (1987) found that negative priming increased when the delay between prime and probe increased from 20 to 520 msec, which suggests that inhibition requires time to develop. Thus, an emphasis on speed may not allow enough

time for the distractors to be inhibited, resulting in facilitation from the initial pa activation. A further difficulty for the episodic retrieval view comes from the finding the finding that negative priming is dependent on stimulus repetition (Strayer & Grison, 1999). Strayer and Grison found reliable negative priming with stimuli that had been repeated throughout the experiment but no evidence of negative priming with novel

stimuli. They argued that episodic retrieval theory should predict as much, and possibl more, negative priming with novel stimuli compared to repeated stimuli because with

novel target stimuli, there is only one episode that can be retrieved. This episode, wh is also the most recent episode and therefore easily accessible, contains conflicting response information and should produce the typical negative priming effect. Thus, in terms of an episodic retrieval view, novel stimuli provide the optimal experimental

context for negative priming to occur. However, this is not the pattern of results foun by Strayer and colleagues (Malley & Strayer, 1995; Strayer & Grison, 1999). The inhibitory account of negative priming can account for these findings if it is assumed that stimulus repetition results in a higher level of activation ofthe internal representations of these stimuli, which would lead to greater response competition. In line with a reactive view of inhibition, increased interference from highly activated

distractors would lead to increased inhibition and consequently, more negative priming. 52

However, perhaps the most significant challenge to episodic retrieval theory comes from recent findings by Neumann, McCloskey and Felio (1999). Using a primed lexical decision task, Neumann et al. (1999) investigated the priming effects produced by attended and ignored words presented in two different languages to bilingual participants. In an initial unilingual experiment (Experiment 1), lexical decisions to target words were faster when the word matched the preceding target word relative to an unrelated word, whereas lexical decisions were slower to target words that had been the ignored distractor word on the preceding display. In the bilingual experiment (Experiment 2), the prime target and distractor words were presented in English, whereas the probe target words were presented in Spanish. Neumann et al. (1999) predicted that if participants knew that the probe target was going to be presented in Spanish, they would be able to globally inhibit the English language following the

prime trial presentation to prevent interference with the current task-relevant languag This should prevent any spread of activation from one language to another and thereby eliminate any potential benefit from a repeated target word. However, no reduction in negative priming would be expected as the distractor word would be implicitly inhibited during prime selection and then globally inhibited as part ofthe English language and should therefore produce delayed responses relative to an unrelated word. In contrast, Neumann et al. (1999) argued that episodic retrieval theory would predict both positive and negative priming on the basis ofthe automatic retrieval ofthe prior processing

episode with the repeated and ignored stimuli respectively. In line with the inhibitionbased hypotheses, Neumann et al. (1999) found a dissociation between the priming effects across languages, with a significant negative priming effect in the ignored

distractor condition but no positive priming effect in the attended repetition conditio

The episodic retrieval account is unable to explain these results, as a target word tha 53

sufficiently similar to an ignored distractor to elicit the retrieval ofthe distractor's incompatible response tag and produce negative priming should also be able to elicit the retrieval of a repeated target word's compatible response tag and produce positive priming. These results seriously challenge the episodic retrieval account of negative priming, and provide some ofthe most direct evidence to date for the role of inhibitory processing in selective attention. In a recent evaluation ofthe current theories of negative priming, Tipper (2001) argues that inhibition and retrieval accounts are not mutually exclusive and that the main difference between the two accounts is the emphasis that is placed on the encoding and retrieval processes involved. According to Tipper (2001), the episodic retrieval accounts have been mainly concerned with the describing the retrieval processes triggered by the probe display and have tended to ignore the selection mechanisms involved in the prime processing episode. In contrast, inhibition accounts more completely describe the mechanisms of selection engaged at the encoding stage ofthe prime display and have failed to consider the role of retrieval in the processing ofthe probe trial. Tipper (2001) presents a revised view ofthe inhibitory mechanisms involved in negative priming, derived, in part, from the Houghton and Tipper (1994) computational model of selective attention. This model will be described in the mechanisms of inhibition section below (see 2.5 Mechanisms of inhibition). The two main points discussed by Tipper (2001) are that inhibition is not necessarily associated with abstract logogen-like representations and is not a simple forward acting process from prime to probe. In contrast, Tipper (2001) argues that the inhibition observed via

negative priming is a bi-directional process involving the inhibition of task-irrelevant

stimulus features during target selection and retrieval processes that are activated whi

interacting with the probe that reinstate the processing involved with the previous prim 54

display. Furthermore, inhibition in this model is reactive such that the greater the activation state ofthe distractor, the greater the inhibitory feedback to the distractor. Recent evidence in support ofthe reactive nature of inhibition has been reported by Grison and Strayer (2001). According to Tipper (2001), inhibition is a flexible process applied to the specific characteristics of an object that is selected against. Moreover,

the retrieval of this inhibitory processing that leads to the negative priming effect. Thi conceptualisation goes some way towards providing a resolution ofthe conflict between inhibition and episodic accounts of negative priming and highlights the importance of considering the retrieval processes involved when attempting to infer inhibitory processes via negative priming effects.

2.5 Mechanisms of inhibition

Although inhibitory processes are incorporated in many cognitive theories, the mechanisms by which this inhibition is achieved are not usually specified beyond general statements concerning the suppression of representations. Two inhibitory mechanisms that have a neurophysiological basis are lateral inhibition and selfinhibition. Lateral inhibition involves the operation of inhibitory connections between units on the same layer of an associative network (Arbuthnott, 1995). Activated units spread inhibition to all other units with which they are negatively associated, the magnitude of which is determined by the unit's own level of activation and the strength

of its inhibitory connections. Lateral inhibition functions to magnify initial difference in activation level, enabling rapid and efficient selection ofthe most highly activated unit (Arbuthnott, 1995). Self-inhibition refers to the suppression of a unit immediately

following its activation to facilitate return to a baseline state of activation as quickl

possible. The main function of self-inhibition is to maintain stability in activation-bas 55

networks, preventing both the build up of extreme levels of activation and the perseveration of one response following activation (Arbuthnott, 1995). In computational models, self-inhibition is often modelled using opponent process units, in which each

unit is attached to one self-excitatory and one self-inhibitory feedback loop. Thus, when activated, each unit generates a combined excitatory-inhibitory feedback signal, which

is able to self-stabilise at any level of activation within the range allowed by the mod One model that incorporates both lateral and self-inhibition mechanisms is the neural network model of selective attention developed by Houghton and Tipper (1994).

2.5.1 The Houghton-Tipper model

The proposed function of inhibitory mechanisms in the Houghton and Tipper

(1994) model is to assist in the efficient selection of target information and to reduce interference from competing distractors. The central feature of this model is the template selection mechanism, which is created according to current goals and task demands. The template contains stimulus features that specify the properties ofthe

current target, such as colour, size, location, shape etc. Perceptual inputs are compared against the template and any that match the features ofthe template receive excitatory

feedback, whereas properties ofthe distractor that fail to match the target specificatio receive inhibitory feedback. Object representations are built up automatically by the visual system, with the various features of an object bound together by excitatory and inhibitory links to form a unified representation. Individual property units generate excitatory and inhibitory feedback onto themselves as well as to other units within an object representation. When a stimulus feature is activated, each property unit sets up combined excitatory-inhibitory feedback signal that is self-stabilising. Feedback from the template selection mechanism breaks the symmetry between excitation and 56

inhibition and shifts the balance so that the activity level ofthe property unit is either enhanced or suppressed. An increase in activation of some subset of property units will

enhance the activation of all property units within an object representation by a proces

of spreading activation. Similarly, the inhibition of a particular feature or property u

will suppress the activity of all the property units within the object representation by analogous spreading inhibition mechanism acting through the linked inhibitory cells. Thus, the properties associated with the target object are enhanced relative to the

distractor object, which allows the coupling of response variables to the target object. The Houghton-Tipper model is able to simulate various interference effects by

allowing separate object assemblies to share property units. The shared units are subjec

to both excitatory and inhibitory influences resulting in an activation level that is hi

than the unshared units ofthe distractor, but lower than the unshared units ofthe target The activation ofthe shared units can then enhance the level of activation ofthe distractor, thus giving rise to greater interference (Houghton & Tipper, 1994). It then follows that the degree of interference should be positively related to the degree of similarity between the target and distractor as the number of shared units between the two representations would be greater, and thus take longer to discriminate. Simulations ofthe model varying the degree of similarity between target and distractor have demonstrated this pattern, with an increase in similarity producing a decreased activation gap between the two object representations, indicating a greater level of interference. This relationship between the degree of similarity between target and distractor and level of interference has been demonstrated in a number of behavioural studies (Bjork & Murray, 1977; Duncan & Humphreys, 1989; Santee & Egeth, 1980). The model can also simulate negative priming effects by way of an inhibitory rebound effect at prime stimulus offset. Although the distractor representation is 57

suppressed relative to the target representation, it is not necessary for the model to

suppress the representation ofthe distractor below resting levels for selection to occu However, once the external excitatory input is terminated at the end ofthe prime display, the property units in a distractor representation receive inhibitory feedback

only, which causes an inhibitory rebound in the overall activation level ofthe distract representation. If the distractor is re-presented as a target during the time that the

representation ofthe distractor is in a suppressed state, it will suffer greater interfe from a novel distractor that has an initial activation advantage and response times will be delayed, thus producing the negative priming effect. Houghton and Tipper (1994) emphasise the role of interference in producing the negative priming effect because in

the absence of an interfering distractor, the representation ofthe inhibited stimulus m rapidly reach an activated state with the onset ofthe probe trial. A number of behavioural studies have shown negative priming to be dependent on the presence of a distractor on the probe trial (Lowe, 1979; Tipper, 1985), however, others have argued

that this is only true if the trials can be easily identified as nonconflict trials (Mo 1994). In this case, a template selection mechanism may not be maintained and negative priming would not be expected to occur. The model can also simulate the complex relationship between the level of distractor interference and the amount of negative priming associated with the distractor. Both positive and negative relationships between interference and negative priming emerge from the model due to the independent excitatory and inhibitory feedback selection mechanisms (see Houghton, Tipper, Weaver, & Shore, 1996). Increasing the inhibitory feedback weights results in the more efficient suppression of the distractor and thus, reduced interference, but an increased inhibitory rebound evidenced as increased negative priming. Thus, in this situation, the model would 58

predict a negative relationship between interference and negative priming. Another

property ofthe model is that the level of inhibition is determined by the activation-st

ofthe distractor, that is, the inhibitory feedback is reactive. Distractors that are mor salient receive greater inhibitory feedback than less salient distractors (Houghton et 1996). This is achieved by the self-regulating feedback mechanism, which enables the strength ofthe inhibition to continually adapt to the strength ofthe input from the distractor (Houghton & Tipper, 1994). Thus, more salient distractors will generate

highly activated representations, and consequently, greater inhibitory feedback resulti in increased interference and increased negative priming. The Houghton-Tipper model provides a computational account ofthe inhibitory selection mechanisms involved in

the interference and negative priming effects, and allows specific predictions regarding the influence of certain task demands to be generated.

2.6 Summary

The constructs of interference and inhibition have become popular as explanatory mechanisms in a number of areas of psychology, including cognitive development and aging, neuropsychology and individual differences research. However, there is still disagreement as to the classification of different inhibitory processes the mechanisms underlying these processes. Consequently, there are a number of tasks and paradigms assumed to tap inhibitory processing, however, these vary greatly depending on the type of inhibition they were designed to measure. The concept of inhibition is attractive because it provides a means to resolve cognitive interference competition. The paradigm that has most often been used to measure this type of inhibition is the negative priming paradigm. This paradigm has been popular because it

is thought to provide a direct index ofthe strength of inhibition applied to a competing 59

stimulus. Having said that, there are alternative accounts of negative priming that do not incorporate inhibitory processes, however, these alternative accounts of negative priming also have difficulty with a number ofthe findings in the negative priming literature. Consequently, the nature ofthe mechanisms underlying the negative priming effect is still open to debate. Recent evidence from Strayer and colleagues and also Neumann and colleagues strongly implicates the involvement of inhibitory processes in the negative priming effect. This is also the most plausible account ofthe relationship between interference and negative priming evidenced in specific populations known to be susceptible to interference. Moreover, the computational model of Houghton and Tipper that incorporates inhibitory mechanisms to resolve interference from response competition

can account for a range of findings in the negative priming literature. Thus, in line wi the Houghton and Tipper model and Nigg's taxonomy of inhibitory processes, the view

taken in this thesis is that the resolution of interference requires the effortful supp of competing responses and that negative priming reflects an inhibitory rebound from

that suppression. Consequently, negative priming will be used as an index of inhibition. In the following chapter, the relationship between inhibition and working memory will be examined more closely by reviewing two models of working memory in detail with respect to how inhibitory processing is explicitly or implicitly incorporated in these models. In addition, a specific framework that focuses on the interaction of inhibition and working memory will be reviewed.

60

Chapter 3 : Models of Working Memory The working memory construct has enjoyed a resurgence of interest in recent

years with a number of independent lines of research examining the structure, locat

and cognitive processes thought to be involved. As a result, several distinct theore positions on working memory have been developed based on evidence from a wide range of sources including cognitive (Baddeley & Hitch, 1974; Conway & Engle, 1994; Cowan, 1988; Engle et al., 1995), neuropsychological (Pennington, Bennetto, McAleer,

& Roberts, 1996; Roberts et al., 1994), developmental (Case et al., 1982; Harnishfege 1995; Harnishfeger & Bjorklund, 1994), aging (Hasher et al., 1991; Salthouse & Babcock, 1991) and computational modeling research (Cohen & Servan-Schreiber, 1992; Kimberg & Farah, 1993). A number of these models are based on specific

populations or specific paradigms of research and so, although relevant, these spec models of working memory may not be the most pertinent for examining the relationship between inhibition and working memory performance. To get a broader

understanding of this relationship, two ofthe most prominent and influential approa

to current working memory research will be reviewed in detail. One of these approach is based on the multi-component model of working memory proposed by Baddeley and Hitch (1974). This approach has been favoured in the United Kingdom and

conceptualises working memory as a tripartite system with a central executive and t

slave systems devoted to storage. Research in this tradition examines the contribut

individual components of working memory to various cognitive activities by explorin performance decrements in dual-task paradigms. The second approach is often described as the North American approach, which typically focuses on individual differences in working memory capacity and how these differences relate to higher-

61

order cognitive abilities. O n e ofthe major contributors to this approach is Randall Engle, who recently proposed a reformulation of his original model of working memory that focuses on the capacity for controlled attention. In a different approach to working memory that has focused more on the relationship of working memory to task performance, Roberts and Pennington (1996) proposed an interactive framework for investigating the relationship between working memory and inhibitory processes. One ofthe aims of this thesis is to examine the interaction between working memory and inhibitory processing proposed in the Roberts and Pennington framework and how such an interaction might be accommodated by the current approaches to working memory. Although these models differ in terms ofthe methodology and populations of interest used in research, some common features have emerged among these theories that appear to be critical to the models of working memory they describe. The most noticeable areas of agreement among these theories is that working memory is limited in capacity and some form of cognitive inhibition is required to prevent irrelevant information from entering and interfering with an upcoming response. The point of divergence between these theories is how these

processes interact to produce the desired response. In line with the review of working memory and inhibition presented in Chapters 1 and 2, the models of Baddeley and Hitch, and Randall Engle will be reviewed according to the structure, function and characteristics of working memory they propose, and the prominence of inhibitory processes in their explication of successful working memory performance. The interactive framework of Roberts and Pennington will also be reviewed as a potential model ofthe interaction between inhibition and working memory processes.

62

3.1 The Multiple-Component Model (Baddeley & Hitch, 1974; Baddeley & Logie, 1999)

Baddeley and Logie (1999) recently defined working memory as " those functional components of cognition that allow humans to comprehend and mentally

represent their immediate environment, to retain information about their immediate pa experience, to support the acquisition of new knowledge, to solve problems, and to formulate, relate, and act on current goals" (p. 28-29). Baddeley and Hitch (1974) proposed a seminal model of working memory that combined the temporary storage of

information with aspects ofthe manipulation and processing of information required to generate a response. This model was derived from studies of healthy adults, children, and brain-damaged individuals, and is able to account for a wide range of empirical findings on working memory.

3.1.1 Structure

The original framework of Baddeley and Hitch (1974) was a multi-component system comprising of a central executive controlling mechanism and two subsidiary systems specialised for the temporary maintenance of information within particular domains. One of these subsystems has been termed the phonological loop and is

dedicated to the processing and storage of verbally coded material. The other subsyst is referred to as the visuospatial sketchpad and is responsible for the storage and

manipulation of visual and/or spatial information. Since the original formulation oft

model, the phonological loop has been further fractionated into a passive phonologica store and an active rehearsal process (Baddeley et al., 1984). The phonological store represents information in a phonological code, which decays over a period of 1 to 2 seconds, whereas the rehearsal process serves to maintain the representations in the 63

phonological store via a process of subvocalization. Similarly, the visuospatial sketchpad has also been fractionated into a visual cache responsible for the storage of visual information and a spatially based rehearsal mechanism termed the inner scribe (Logie, 1995). Evidence for the independence ofthe two subsystems comes from dualtask experiments that show task specific interference effects (Baddeley, 1986). For example, Logie, Zucco and Baddeley (1990) reported a double dissociation in which a

visual imaging task significantly interfered with participants' visual span performance

but minimally with verbal span, whereas a verbal addition task significantly interfered with participants' verbal span but minimally with visual span. They concluded that the phonological loop and the visuospatial sketchpad were semi-independent specialised mechanisms that could act independently or concurrently according to task demands. The central executive component ofthe model is a limited-capacity attentional mechanism responsible for the control and coordination ofthe working memory system. This component has often been criticised as little more than a homunculus. In response to this, Baddeley (1996) presented four lines of research that illustrated some ofthe

specific functions served by the central executive. The first of these was concerned wi the capacity to coordinate the two storage subsystems, which has typically been examined by means of dual-task performance. Evidence from patients with Alzheimer's Disease (AD) has shown that dual-task performance is impaired in these patients relative to age-matched controls, even though the constituent tasks were equated for difficulty across groups (Baddeley, Logie, Bressi, & Delia Sala, 1986). This suggests that the capacity to coordinate two separate tasks is a separable feature ofthe central executive and that this ability is specifically impaired in AD patients. The second

component function concerns the ability to switch attention, which arose out of researc examining random generation performance. The random generation task involves the 64

rapid generation of random sequences of letters or numbers, which necessarily requires the avoidance of stereotypical counting responses or the recital ofthe alphabet. Baddeley, Emslie, Kolodny, Duncan (1998a) argued that in order to prevent the

production of stereotypical responses, the retrieval plans used to generate the response must be constantly switched. In support of this, they found random key-pressing to be

significantly impaired by a verbal equivalent ofthe Trails Test in which participants ha to constantly switch between producing consecutive letter and consecutive number sequences, but unaffected by reciting the alphabet or counting. Baddeley (1996)

concluded that the capacity to switch attention between retrieval strategies, as reflect in random generation, is one ofthe component functions ofthe central executive. The third component ofthe central executive proposed by Baddeley (1996) is

the capacity for focused selective attention. Baddeley (1996) reports a series of studies that he and his colleagues conducted to examine the working memory deficit often reported in the elderly. Elderly and middle-aged participants completed a selective

attention task in which they were required to respond to a specific stimulus as rapidly a possible, both with and without the presence of irrelevant stimuli. They found that the elderly participants were significantly slower than the middle-aged participants to respond to the target stimulus in the presence of irrelevant stimuli, and that this difference remained when the influence of IQ and speed of processing were partialled out. They concluded that neither a simple slowing in general speed of processing nor a

decline in fluid intelligence could account for age differences in the capacity to ignore irrelevant stimuli. Instead, they suggested that their results were consistent with the

that age limits the capacity for utilising inhibition to focus attention and limit distr and that this may be an important function ofthe central executive. The final executive component proposed by Baddeley (1996), is the capacity for the temporary activation of 65

long-term memory. In a seminal paper, Hulme, Maughan and B r o w n (1991) demonstrated that memory span for words was better than memory span for nonwords, and argued that this provided evidence of a long-term memory contribution to shortterm memory performance. In terms ofthe multiple-component model, findings such as

these suggest that the phonological loop may have an interactive relationship with lo term memory. One way of conceptualising the relationship between short-term and

long-term memory may be through a central executive system that is responsible for the encoding and retrieval of information from both the slave systems and the temporarily activated components of long-term memory (Baddeley, 1996). In support of this, research by Engle and colleagues has demonstrated that individual differences in working memory capacity predict performance on a number of tasks thought to rely on long-term memory activation such as the fan effect (Cantor & Engle, 1993) and semantic category generation (Rosen & Engle, 1997). Baddeley (1996) does not regard these functions as the only ones served by the central executive and remains open on issue of whether they are independent but interacting control processes or reflect a unified system with multiple functions. Baddeley (2000a) recently proposed a fourth component to the working memory model called the episodic buffer. The episodic buffer is assumed to be a limitedcapacity temporary storage system that is capable of integrating information from the subsidiary systems into a temporary representation. The episodic buffer provides an interface between the subsidiary systems and long term memory and has the capacity to store information in a common multi-dimensional code. The central executive is assumed to control the episodic buffer and can influence the content ofthe buffer by attending to a given source of information. Information is integrated from various sources across space and time into coherent episodes, which can be retrieved by the 66

central executive through conscious awareness. The buffer provides a mechanism for

modeling the environment and creating new cognitive representations and as such, play an important role in long-term episodic learning (Baddeley, 2000a). Support for the existence ofthe episodic component comes from a neuroimaging study by Prabhakaran, Narayanan, Zhao and Gabrieli (2000) which demonstrated differential activation

patterns for the retention of verbal and spatial information held in an integrated fo compared to when it was unintegrated. The episodic buffer shifts the emphasis from

isolating the subsidiary systems to the process of integrating information and provid crucial link between long-term memory and the working memory system.

3.1.2 Function

Baddeley and Logie (1999) believe that any complex cognitive activity requires the involvement of multiple components of working memory and the coordination of information among them. One area that has received considerable investigation is the contribution ofthe phonological loop and central executive components of working memory to language comprehension (see Gathercole & Baddeley, 1993). Baddeley and Logie (1999) argue that the capacity to comprehend a particular passage will be determined both by the existing long-term memory representations and the capacity of the central executive to activate and combine representations into a coherent form. There is also considerable evidence that the phonological loop component plays a crucial role in vocabulary acquisition, particularly in relation to foreign-language learning (Baddeley, Gathercole, & Papagno, 1998b; Papagno, Valentine, & Baddeley, 1991; Papagno & Vallar, 1995). For example, Papagno et al. (1991) found that

articulatory suppression disrupted the learning of foreign words in a paired-associat

67

task, but not the learning of paired-associates presented in the participants' native language. Another complex activity that involves different components of working memory is mental arithmetic. Logie, Gilhooly and Wynn (1994) examined the role of the phonological loop, visuospatial and central executive components of working memory in mental arithmetic using a dual-task procedure in which a cumulative

addition task was paired with either articulatory suppression, spatial tapping or ran

generation. Articulatory suppression and random generation both produced a significan increase in the number of errors, whereas spatial tapping did not, suggesting a role

the phonological loop and central executive components in mental arithmetic. However, the error responses produced by participants were close to the correct total, which

suggests that participants could still generate reasonable approximations, possibly b some form of estimation strategy implemented by the central executive. Logie et al. (1994) argued that mental arithmetic was reliant on both components of working memory; the phonological loop for the maintenance of accuracy through subvocal rehearsal and the storage of partial solutions, and the central executive for the application of calculation procedures and estimation strategies. Although the subcomponents of working memory are demonstrably involved in a range of cognitive

tasks, it is beyond the scope of this thesis to review them all. Nevertheless, the st discussed above highlight the value ofthe multiple-component model in explaining cognitive task performance.

3.1.3 Capacity limitations

The subcomponents ofthe working memory model each have individual

constraints on capacity associated with the particular functions they perform (Badde 68

& Logie, 1999). The capacity ofthe phonological loop is also constrained by the

amount of information that can be rehearsed in approximately 1.5 seconds. This is ofte examined by means ofthe word-length effect, which refers to the finding that memory span for words is inversely related to the spoken duration ofthe to-be-recalled items (Baddeley, Thomson, & Buchanan, 1975). Consequently, individuals who are able to rehearsal more efficiently are likely to retain more items in the phonological loop. recent review focusing on the development ofthe phonological loop, Gathercole (1999) outlined a number of cognitive mechanisms that have been linked to developmental increases in verbal short-term memory capacity including the perceptual analysis of information, the construction and maintenance of a memory trace, order memory, and subvocal rehearsal. Thus, limitations in these components during development may

impact on an individual's capacity for the short-term storage of verbal information in adulthood. Research examining the constraints of visuospatial working memory suggests a

potential dissociation between the capacity for retaining visual patterns and that for

retaining sequences of movements. Evidence in support of this distinction comes from a developmental study by Logie and Pearson (1997), in which children of various ages

were tested on their recall of matrix patterns and spatial pointing sequences. They f that memory span for matrix patterns improved more rapidly with age than span for spatial sequences, suggesting the capacity for retaining different types of visual information develops at different rates (Logie & Pearson, 1997). However, this may be linked to the development of phonological recoding strategies in the older children (Pickering, 2001). The nature ofthe limitations on memory for spatial information are yet to be examined in detail, however, research examining visual pattern memory in

adults has demonstrated a one-item recency effect in recognition memory for a sequence 69

of matrix patterns (Phillips & Christie, 1977), which suggests that visual m e m o r y m a y

be limited to a single item. However, the mechanisms underlying individual differences in visuospatial memory capacity are still open to further exploration. The central executive component ofthe working memory system is also assumed to be limited in capacity, however, the nature of these constraints is not

explicitly clear. Baddeley (1996) believes that the central executive can be fractiona into component processes that are necessary for executive control. However to date,

Baddeley (1999) is undecided as to whether the central executive will ultimately prove to be a hierarchy of processes with a dominant controller or a range of independent control processes. As in the case ofthe phonological loop and visuospatial sketchpad, each of these processes may have limitations commensurate with the functions they perform. Baddeley (1999) also leaves open the possibility that many of these subprocesses depend on a common underlying mechanism such as excitation or inhibition.

3.1.4 Control mechanisms

The working memory system is controlled by the central executive, which is

typically regarded as an attentional system responsible for the integration of informa and the control of action (Baddeley, 1993). The initial formulation ofthe central executive was strongly influenced by the model of attentional control proposed by Norman and Shallice (1986), which was described by Baddeley (1986) as an adequate approximation of central executive functioning. Norman and Shallice's model of attentional control assumes that two complementary processes operate in the selection and control of action. The basic mechanism is termed contention scheduling, which is

thought to be able to control routine activities automatically, without conscious cont 70

or attentional resources (Norman & Shallice, 1986). In nonroutine situations requiring novel or difficult actions, the contention scheduling mechanism is modulated by the deliberate, conscious control ofthe supervisory attentional system (SAS) (Norman & Shallice, 1986; Shallice, Burgess, Schon, & Baxter, 1989). The model is based on the operation of a series of self-contained, well-leamed action and thought sequences termed schemata. A schema can be activated by welllearned triggers, either from the perceptual system or the output of recently active schemata, and is selected once the activation level reaches threshold. The contention scheduling mechanism prevents schemata from conflicting and competing for the same cognitive resource by means of a lateral inhibitory mechanism (Shallice & Burgess, 1991). However, conflicts between potential action schemata are inevitable, and so a conflict resolution procedure is necessary. The SAS performs this function by

modulating the activation level ofthe schemata, thus biasing their probability of bein selected (Shallice & Burgess, 1991). The higher-order processes ofthe SAS are

implicated when the conscious control of action is required. The operation ofthe SAS i thought to be necessary for appropriate behaviour in situations that involve planning decision making, error correction, contain novel sequences of actions or technically difficult actions, and when the overcoming of a strong habitual response is required (Norman & Shallice, 1986). An impairment ofthe SAS should then lead to difficulties

in these situations and on tasks that appear to make strong demands on the functions o the supervisory system. Shallice (1982; 1988) documented a number of frontal lobe patients who demonstrated deficits in cognitive processing consistent with those postulated to result from impairment ofthe SAS. These patients were described as having a deficit in executive processing.

71

3.1.5 Inhibitory processing

Although Baddeley has not directly proposed a role for inhibitory processing in the working memory model, he does not deny the existence or importance of such a mechanism. In his discussion ofthe selective attention function ofthe central executive, Baddeley (1996) reports a study in which elderly participants were slower to respond to a target when they were required to ignore irrelevant stimuli within the same dimension as the target. This effect was still present when differences in IQ and speed of processing were controlled for. Responses for the elderly participants were also slowed by the presence of irrelevant stimuli in a different sensory dimension to the target, however this effect was eliminated when IQ differences were partialled out. Baddeley

(1996) concluded that the results were consistent with theories that argue that age limi the capacity for inhibition. However, the fact that the effect did not occur when the irrelevant stimuli were presented in a different sensory dimension, led Baddeley (1996) to argue against a general reduction in inhibition. He speculated that inhibition might depend on an attentional distribution represented by a bell-shaped curve, with the

features specifying the target stimulus at its centre. Stimuli with similar characteristi to a target will be closer to the focal point ofthe distribution and hence, receive more attention and more efficient processing. In contrast, stimuli falling outside ofthe distribution will be ignored, and those falling on the edges will require some processing before they are rejected (Baddeley, 1996). Baddeley (1996) suggested that age may lead

to a less highly peaked distribution of attention. Consequently, irrelevant stimuli withi the same sensory dimension are more likely to fall within the broader attentional distribution of older participants and therefore receive more processing. Furthermore,

irrelevant stimuli presented in a different sensory dimension will fall outside the focu of attention and hence, be ignored (Baddeley, 1996). Baddeley (1996) emphasises that 72

his view ofthe inhibition process is purely speculative, however, it does point to the importance of inhibition in the central executive component ofthe working memory model.

3.2 The Controlled Attention Model (Engle, Kane & Tuholski, 1999a

Engle, Kane and Tuholski (1999a) define working memory as a system comprising of long-term memory traces that are active above threshold, the processes

necessary to achieve and maintain that activation, and controlled attention. This mod has been derived from the North American tradition of examining working memory

from the perspective of individual differences research. Much of Engle's work has been concerned with explaining individual differences in working memory capacity and how

these differences relate to higher-level cognitive activities. In attempting to under

these differences, Engle has examined the performance of extreme groups of individual who score in the upper and lower quartile on a variety working memory span tasks such as the reading span and operation span tasks.

3.2.1 Structure

In his most recent formulation of working memory, Engle et al.( 1999a) proposes a system that consists ofthe contents of short-term memory plus controlled attention processes. This model was strongly influenced by the embedded-processes model of Cowan (1988; 1993), which described a single memory system with different elements

at various levels of activation. In line with Cowan (1988), Engle et al. (1999a) descr short-term memory as the portion of long-term memory activated above resting

baseline, with representations in the form of phonological traces, visual traces or an

other trace that may be associated with perception, emotion, and thought. All traces a 73

subject to the same principles of forgetting and interference, with loss of activation caused by decay and/or inhibition. A small number of these traces are maintained in the focus of attention through controlled attention processes that increase the activation particular traces according to current task goals. Controlled attention is a limited

capacity mechanism responsible for activating representations to either bring them into

focus or maintain them in focus, particularly in the face of interference or distraction (Engle et al., 1999a). The activation of long-term memory traces is achieved through controlled retrieval. When necessary, the controlled attention mechanism can also be used to dampen activation of representations through inhibition. The controlled attention mechanism is conceptually similar to the central executive of Baddeley and Hitch's (1974) multiple-component model. Engle, Tuholski, Laughlin & Conway (1999b) performed an analysis ofthe unique and shared variance across a range of tasks thought to reflect short-term memory and working memory and examined the'underlying factor structure of that variance. Engle et al. (1999b) examined the performance of 133 participants on a number of working memory tasks including reading span, operation span and counting span, and a range of short-term memory tasks including forward and backward word span with dissimilar words and forward word span with similar words. In addition to the memory tasks, participants performed the Raven's Standard Progressive Matrices and Cattell Culture Fair Test as measures of fluid intelligence (gF). A two-factor confirmatory factor analysis model showed that the three working memory tasks were linked to one latent variable, whereas the three short-term memory tasks were linked to a second latent variable. Furthermore, the two-factor model provided a significantly better fit

the data than a one-factor model, providing support for the conceptualisation of workin memory and short-term memory as separable constructs. In addition, structural equation 74

modeling showed that the latent working m e m o r y variable was significantly related to general fluid intelligence, whereas no connection was required between the short-term memory variable and gF. Engle and his colleagues (1999a; 1999b) have argued that the capacity for controlled attention is strongly related to fluid intelligence. They reasoned that any shared variance between working memory and short-term memory tasks should reflect the short-term storage component common to both, while any residual variance in the working memory variable should reflect the controlled attention component of working memory. In order to examine the relationship between controlled attention and fluid intelligence, Engle et al. (1999b) partialled out the variance common to both working memory and short-term memory constructs and found a significant correlation between the residual working memory variance and fluid intelligence. Moreover, the link between the short-term memory residual and gF was not significant. Engle et al. (1999b) concluded that short-term memory and working memory are highly related but

separable constructs and that the primary factor contributing to the relationship betw measures of working memory and fluid intelligence is controlled attention.

3.2.2 Function

Individual differences studies have implicated working memory in a range of complex cognitive activities. Engle et al. (1999a) argues that working memory span

performance correlates significantly with higher-order ability scores even when the ty of processing involved in the tasks does not match. Turner and Engle (1989) compared the correlation of different span measures with reading comprehension to examine whether the background element ofthe complex span task must be commensurate with

the ability of interest in order for the span score to predict performance on the task. 75

They found that scores on an operation span task, that involved the calculation of simple arithmetic and the storage of words or digits, correlated with measures of reading

comprehension as well as reading span scores despite requiring different skills to thos involved in reading. Furthermore, Engle, Cantor and Carullo (1992) measured the time it took participants to perform the various elements ofthe reading span and operation

span tasks and found that differences in the processing time measure did not account fo the relationship between span and reading comprehension. Taken together, these results suggest that the relationship between measures of working memory capacity and higherorder cognitive tasks is not a result ofthe specific processing components ofthe working memory tasks. Engle et al. (1999a) argue that these results and similar findings by Conway and Engle (1996) support their view that the critical element of working memory tasks that

drives the correlation with higher-order cognitive activities is the capacity for contr

attention. Further support for this view comes from Engle et al.'s (1999b) latent varia analysis discussed previously. Engle et al. (1999b) examined the relationship between the latent working memory and short-term memory constructs and two measures of attainment. They demonstrated that the latent working memory variable accounted for variance in the verbal and quantitative SAT scores above and beyond that explained by the short-term memory construct. According to Engle et al.'s (1999a) conceptualisation

of working memory, this reflects the contribution ofthe controlled attention component. Furthermore, the latent variable analysis and the results of Turner and Engle (1989) suggest that this component ofthe working memory system is unitary and domain-free.

76

3.2.3 Capacity limitations

Engle et al. (1999a) believe individual differences in working memory capacity reflect differences in the capability for controlled attention. Accordingly, these differences will only be evident in situations that demand controlled processing and should not arise when a task can be performed automatically. Support for this view comes from a recent study by Tuholski, Engle and Bay lis (2001) in which participants who scored high or low on a working memory span task, performed an enumeration

task that involved the presentation of n objects for them to count. Tuholski et al. (20 argued that when n is from one to four, enumeration is presumed to occur through an

automatic subitizing process. When n is greater than four, participants are presumed to invoke a controlled counting process that involves keeping track ofthe items counted, planning where to move attention next and avoiding recounting by inhibiting previously viewed locations. Tuholski et al. (2001) predicted that participants with less working memory capacity would have steeper reaction time slopes on the counting trials ofthe enumeration task than participants with greater working memory capacities, whereas reaction time slopes on the subitizing section ofthe tasks should not differ between

groups. The results were exactly as predicted. Tuholski et al. (2001) concluded that th performance of high- and low-working memory capacity participants differed on the enumeration task only to the extent that the task required controlled processing. Engle et al. (1992) believe that individuals differ in the total level of activation available to their system and that complex span tasks measure the amount of activation available to an individual on a moment-to-moment basis. Complex span tasks require

the individual to continually switch attention between the processing component and th storage item until recall is required (Towse, Hitch, & Hutton, 1998; Towse, Hitch, & Hutton, 2000). Engle et al. (1992) argued that performance on the span tasks was a 77

reflection ofthe activation available for maintaining the m e m o r y elements in the focus of attention while switching attention between activities. This attention-switching feature is a critical component ofthe working memory span task (Cantor & Engle, 1993). The presentation of each new operation requires attention to be shifted away from the to-be-remembered word. As a result, activation ofthe word drops below the working memory threshold and must be re-accessed from long-term memory during recall (Cantor & Engle, 1993). If there is interference from competing information, the retrieval process may take longer and is more prone to error. According to Engle et al. (1999b) the number of traces that are active above threshold at a given time is limited

by the decay rate and the ability to perform the processes that maintain activation abo threshold (i.e. rehearsal rate). Another limitation is in the number of memory elements

that can be maintained in the current focus of attention. Engle et al. (1999b) argue tha performance on working memory tasks reflects both limitations, but that short-term memory performance primarily reflects the limitations associated with decay and rehearsal rate.

3.2.4 Control mechanisms

In the present model, the regulation of working memory is achieved through the controlled attention component. According to Engle et al. (1999a) controlled attention required to maintain temporary goals in working memory, particularly in the face of interference or distraction. Controlled processing is necessary to focus, divide and switch attention and prevent inappropriate actions through inhibition. Controlled attention is also required when conflict among actions or response competition must be resolved, when task irrelevant information must be inhibited, and for error monitoring and correction (Engle et al., 1999a). Engle et al. (1999b) argue that individuals will 78

differ in their reliance on the controlled attention component and that tasks that are working memory tasks for some individuals will be short-term memory tasks for others. This is dependent on an individual's vulnerability to interference and the extent to which the procedures for achieving and maintaining activation are routinized. Engle et al. (1999a) believe that the functions ofthe controlled attention mechanism are mediated by the prefrontal cortex. Patients with prefrontal cortex lesions demonstrate difficulty with vigilance tasks, increased susceptibility to distractors, and make more

intrusion errors from prior lists in a proactive interference task than controls (see E et al., 1999a, for further detail). Thus, the prefrontal cortex appears to be important focusing and maintaining attention. Engle et al. (1999a) suggest that individual

differences in controlled attention are likely to be mediated by individual differences prefrontal cortex functioning. The operation ofthe controlled attention mechanism is

not well specified, and Engle et al. (1999a) rely on similar control mechanisms, such as

the central executive and the supervisory attentional system, to describe this componen ofthe model. The major difference between these conceptualisations is that in contrast to the central executive ofthe multiple-component model, the controlled attention mechanism is assumed to be unitary in nature and dependent on a limited amount of activation that can be distributed according to task demands (Conway & Engle, 1996; Engle, 2002; Rosen & Engle; 1997).

3.2.5 Inhibitory processing

Following the influential study by Conway and Engle (1994) discussed in Chapter 1 (1.4.1 Capacity limitations), which demonstrated that individual differences in working memory capacity correspond to differences in the ability to suppress irrelevant information, inhibition has taken a prominent role in the evolution ofthe 79

controlled attention model. Engle (1996) reformulated his original general capacity

theory into the inhibition-resource hypothesis, and argued that the act of inhibit

required limited attentional resources. According to this view, individual differe

inhibitory processing are the result of differences in the capacity for controlled intentional processing rather than automatic aspects of memory (Engle, 1996). The

obvious prediction from this view is that if inhibition requires attentional resou

inhibition should be more difficult when attention is required by another secondar Engle, Conway, Tuholski and Shisler (1995) examined this by having participants perform a letter-naming task and a recall task simultaneously. An example ofthe procedure used in this study is presented in Figure 3.1. ##**

LoadO

Load 1

Load 2

{ { {

B

^ ^

Inhibition Trial

^ "i

Control Trial

4

Inhibition Trial

Y

B room G

z

sfe if. *k -J*k jfe ^4, .,-.

x

u

hand 0.. M A A G fact x

o

Load 3 z

Load 4

A

{

u

time Dy ^

A D

recall

Figure 3.1: Diagram adapted from Engle, Conway, Tuholski and Shisler (1995) showing the procedure used in their letter-naming task.

80

Each trial in the letter-naming task consisted of a prime display followed by a probe display. One third ofthe prime-probe pairs presented in the letter-naming task formed negative priming trials in which the ignored distractor letter from the prime display became the target letter in the subsequent probe display. Two thirds were

control trials in which there was no relationship between the letters in the prime di and the subsequent probe display. The difference in naming latencies between these

trials provided a measure ofthe negative priming effect, which was taken as an index o

inhibitory processing. Sequences of five prime-probe pairs were constructed. Followin

each prime-probe pair, a word was presented that was to be remembered for recall at th end ofthe sequence. The first prime-probe pair was assumed to be under a memory load of zero as the first word was yet to be presented. The second prime-probe pair that

followed the presentation ofthe first word was assumed to be performed under a load o

one because the participant had one item in memory. In a similar fashion, the last pri

probe pair was performed under a load of four as the participant now had four items i memory. Engle et al. (1995) argued that if inhibition was attention demanding, the negative priming effect should be reduced by an increase in memory load. They found a significant negative priming effect when there was no memory load, which as expected, decreased as memory load increased from one to four items. Engle et al. (1995)

concluded that inhibition was a product of controlled resources and that the magnitud ofthe negative priming effect was dependent on the momentary resources available to the individual. Conway, Tuholski, Shisler and Engle (1999) extended this research to examine the effects of a verbal and a nonverbal memory load on negative priming. Using a similar paradigm, they presented participants with letter-naming trials interspersed either words (verbal memory items) or random polygons (nonverbal memory items) as 81

the to-be-remembered items. At the end of each trial, participants were presented with either a word in the verbal condition or a random polygon in the nonverbal condition and were required to decide whether the item matched one ofthe four items presented in the trial. Conway et al. (1999) found significant negative priming effects in both the verbal and nonverbal memory load experiments. The size ofthe negative priming effect did not interact with memory load. However, planned comparisons revealed a small but reliable negative priming effect when memory load was zero that was not observed under any other load condition (Conway et al., 1999). This pattern was observed with both verbal and nonverbal memory loads. In addition, Conway et al. (1999) examined the performance of participants classified as high or low in working memory capacity and found that high-span individuals demonstrated significant negative priming when the memory load was zero, whereas low-span individuals did not demonstrate negative priming at any level of load. Again, this pattern was observed in both the verbal and nonverbal memory load experiments. Conway et al. (1999) concluded that the processes that contribute to the negative priming effect are resource dependent and that the resource pool is domain free. Furthermore, individual differences in the amount of resources available result in individual differences in the negative priming effect. Conway et al. (1999) argued that inhibition was the mechanism responsible for resolving competition from distracting information and that inhibition was resource dependent. Moreover, Conway et al. (1999) argued that working memory capacity

drives the ability to resolve interference and that individuals with high working memo

capacity are more likely to inhibit distracting information. According to this framewo

the ability to inhibit irrelevant information corresponds to the capacity for controll attention.

82

3.3 The Interactive Framework (Roberts & Pennington, 1996) Recent research into prefrontal executive functions has identified the importance of working memory processes and inhibition in determining action. Roberts, Hager and Heron (1994) analysed a range of tasks sensitive to prefrontal functioning and concluded that most involved an underlying competition between response alternatives. Another common requirement of prefrontal tasks identified by Roberts et al. (1994) is the involvement of some form of working memory processes to keep task-relevant information in mind and to use that information to generate an appropriate response.

This analysis ofthe processes involved in prefrontal tasks is consistent with the grow

body of research in other areas of cognition that has also made the association betwee working memory processes and response inhibition (Bjorklund & Harnishfeger, 1990; Harnishfeger & Bjorklund, 1994; Hasher & Zacks, 1988). The nature of this relationship, however, is poorly understood. Roberts et al. (1994) suggested that the ability to inhibit a prepotent response was dependent on the activation of working memory processes. They believed that working memory activation influenced the extent of inhibitory spread to other responses, such that when working memory processes are appropriately activated and maintained, the inhibition of competing response

alternatives occurs by default (Roberts et al., 1994). This is consistent with evidence from computational models (Cohen & Servan-Schreiber, 1992; Kimberg & Farah, 1993) that have shown that disrupting specific parameters that that correspond to working

memory processes produces patterns of performance typical of patients characterised as having difficulties with inhibition. Roberts and Pennington (1996) proposed an interactive framework for

describing prefrontal cognitive processes. In this framework, working memory refers to the ability to maintain and manipulate short-term information required for guiding 83

current actions appropriately. Roberts & Pennington (1996) believe that working memory involves storage, computation and attentional activation. They refer to three

characteristics of working memory, which delineate different aspects of its functioning One characteristic is capacity, which is used in this framework to refer to concurrent storage and processing. Many researchers have focused on capacity as an index of individual differences in working memory and as a result, a number of complex span tasks have been developed as measures of working memory capacity (Daneman & Carpenter, 1980; La Pointe & Engle, 1990; Turner & Engle, 1989). The maintenance of information over time is another characteristic of working memory that is often measured using delayed search tasks (Roberts et al., 1994; Roberts & Pennington, 1996). Useful information about the location of an object must be maintained over

various time intervals until a search for that object is initiated. The third character working memory described by Roberts and Pennington (1996), is the level of momentto-moment activation at any particular point in time. This is important in tasks with

strong prepotent responses that must be avoided, such as the Stroop task and antisaccad

task. Thus, specific tasks are associated with the different characteristics of working memory described in this framework. Whether these features of working memory interact with one another or are separable processes is yet to be established (Roberts al., 1994; Roberts & Pennington, 1996). Roberts and Pennington (1996) describe inhibition as the process involved in preventing an action from occurring or interfering with upcoming action. Although Roberts and Pennington (1996) distinguish this type of inhibition from other forms of inhibition, such as the ability to stop an action that is already in progress and involuntary cognitive inhibition, they accept that these inhibitory processes may be

related. They argue that prepotent actions require inhibition to be suppressed and that 84

the prepotency strengthens, higher degrees of inhibition will be required. According to this framework, inhibition occurs as a by-product ofthe working memory processes involved in generating appropriate responses. The prediction made from the proposal of Roberts and Pennington (1996) is that increasing the working memory demands ofthe task will compromise working memory performance, resulting in decreased inhibition and an increased probability of making an incorrect response. Roberts et al. (1994) investigated this hypothesis by examining the performance of participants on the antisaccade task, which involves a strong reflexive tendency to make a saccade in the direction of a briefly presented cue. Successful performance on this task requires individuals to inhibit this prepotent saccade and make an antisaccade in the opposite

direction to identify a target stimulus. Participants performed the antisaccade task a and concurrently with one of three secondary tasks that varied in the demands they placed on working memory. As predicted, the results showed an increase in the

proportion of reflexive saccades (the incorrect prepotent response) when the concurren secondary task placed more demand on working memory (Roberts et al., 1994). In addition, the antisaccades took longer to initiate than reflexive saccades and were further slowed by the introduction ofthe concurrent task. This supported the idea that the antisaccade task employed the controlled processes of working memory and that increasing the load on working memory decreased the resources available to generate

the antisaccade. Roberts et al. (1994) concluded that successful inhibition of prepote responses was a function ofthe strength ofthe prepotency, the working memory resources available to an individual, and the working memory demands ofthe task. Roberts et al. (1994) suggested that the results from the antisaccade task provided support for the interaction of working memory characteristics, as taxing capacity with the secondary task presumably reduced the level of moment-to-moment 85

activation engaged in preventing a reflexive saccade. However, they failed to find any correlation between performance on the antisaccade task and two measures of working

memory capacity (Roberts et al., 1994). To account for this finding they suggested that the antisaccade task and capacity tasks may be assessing different aspects of working memory, which do not correlate in a normal population but presumably affect each

other when one is impaired or engaged in other activities (Roberts et al., 1994). These conclusions do not appear to be consistent with the proposed interaction between working memory resources, the demand on working memory and inhibitory processes. Furthermore, while Kane, Bleckley, Conway and Engle (2001) did find a relationship between working memory and performance on the antisaccade task such that participants with low-working memory spans were slower and less accurate than highspan participants on the antisaccade task, Miyake et al. (2000) did not find any correlation between performance on an antisaccade task and a measure of working memory capacity. These findings indicate that the relationship between working memory and inhibition is anything but clear. As Roberts and Pennington (1996) suggest, more research directly examining the relationship between these processes is required.

3.4 Overview of thesis Converging lines of evidence from developmental, individual differences and neuropsychological research suggest that there is a relationship between working memory and inhibition. However, the nature ofthe relationship is unclear. As reviewed in Chapter 2, some researchers (e.g. Hasher & Zacks) have argued that the capacity of working memory is limited by the efficiency of inhibitory mechanisms, whereas others

have argued that working memory capacity constrains the ability to inhibit distracting 86

information (e.g. C o n w a y & Engle). There is also discrepancy over the type of

inhibition referred to, whether it is cognitive or behavioural, voluntary or involunta anticipatory or reactive. Although many current models of working memory incorporate inhibitory processes, this is often done superficially without specifying how these processes relate to other components of working memory or the role they play in complex cognitive activities. Baddeley acknowledges that inhibitory processing may be

an important function ofthe central executive but as yet, has not specified this funct in any detail. Engle incorporates inhibition as a central process in the controlled attention model of working memory and argues that the capacity of working memory constrains the ability to inhibit irrelevant information. However, direct evidence in

support of this claim is limited and more research is needed to specify the nature ofth relationship in greater detail. The framework of Roberts and Pennington provides one way of conceptualising the relationship between inhibition and working memory. According to this framework, successful action results from an interaction among an individual's working memory resources, the working memory demands ofthe task, and the strength ofthe competing responses. The prediction made from this framework is that increasing the working memory demands ofthe task will decrease inhibition and therefore increase the probability of making an incorrect response. Although this framework has received some support from the antisaccade study of Roberts et al. (1994), most studies do not systematically vary the demand on working memory within the same task. Furthermore,

Roberts and Pennington (1994) describe inhibition in terms of preventing an action from occurring or interfering with upcoming action and distinguish this from other forms of inhibition. However, converging evidence suggests that the relationship between working memory and inhibition may be more general and that an individual's ability to 87

inhibit irrelevant information is related to their working m e m o r y performance. A s such, we might expect the relationship to be evident across a range of working memory tasks and forms of inhibitory processing. Conway et al. (1999) provided some support for the generality ofthe interactive relationship with the demonstration that negative priming was eliminated with the addition of a verbal and a non-verbal memory load. This is consistent with the predictions ofthe Roberts and Pennington (1994) framework, that increasing the load on working memory compromised the ability to inhibit the irrelevant distractors, resulting in less negative priming. Thus, the relationship appears to be applicable to cognitive inhibition associated with negative priming effects. However, Conway et al. (1999) and Engle et al. (1995) manipulated working memory load by interspersing memory items between the negative priming stimuli and found that negative priming was eliminated with a memory load of one item. In effect, this creates a dual task situation in which participants attend to naming the negative priming stimuli, then switch attention to encode the memory item, then switch again etc. An alternative

explanation ofthe results found by Conway et al. (1999) and Engle et al. (1995) is tha

the processes involved in negative priming are sensitive to task switching rather than cognitive load per se. This would explain the elimination of negative priming with a cognitive load of one item, which is not typically regarded as cognitively demanding. would also provide an alternative account ofthe finding that negative priming was eliminated with both a verbal and a non-verbal memory load. Task switching is known to be cognitively demanding (Allport et al., 1994; Rogers & Monsell, 1995) and may demand resources that would otherwise be available to devote to the negative priming

task. Thus, the results of Conway et al. (1999) and Engle et al. (1995) may not reflec

88

domain general relationship between working m e m o r y and inhibition, but m a y be specific to their manipulation of cognitive load. The main objective of this thesis is to examine the predictions ofthe Roberts and Pennington (1994) framework across three different experimental paradigms. These paradigms were designed to examine the influence of different working memory loads on the inhibitory processes involved in the interference and negative priming effects associated with selective attention (see Chapter 2). This will provide a greater understanding ofthe relationship between working memory and inhibitory processing in cognitive tasks and determine whether the relationship is domain general or dependent on the manipulation of cognitive load involved. Each experimental chapter examines a different paradigm and in accordance with the Roberts and Pennington framework, the main prediction in all experiments is that increasing the load on working memory will compromise working memory performance and decrease the ability to inhibit competing responses. Consistent with an inhibitory view of negative priming and Houghton and Tipper's model of selective attention, this would be evidenced by an increase in the interference encountered from an irrelevant distractor and a decrease in the negative priming effect. Although it was not the original intention of this thesis, the following studies will also be examined in light of Lavie's (1995; Lavie & Fox, 2000) perceptual load hypothesis of selective attention. As reviewed in Chapter 2, Lavie (1995; Lavie & Fox,

2000) clearly demonstrated that the ability to ignore irrelevant information is direct related to the load in the processing of relevant information. However, Engle et al. (1999a) argues that the ability to ignore irrelevant information is constrained by working memory capacity and demonstrates that the processes involved in selective attention are dependent on the cognitive load in the task. This leads to competing 89

predictions with regards to interference effects under different load conditions, which will be elaborated on in the relevant experiments. A comparison ofthe effects of memory load versus perceptual load on irrelevant distractor processing will provide an initial attempt at dissociating these load effects and resolving some ofthe conflicting findings in the literature.

90

Chapter 4 : The Stroop Task 4.1 Experiment 1 The purpose of Experiment 1 was to examine the predictions ofthe Roberts and Pennington (1996) framework in relation to the inhibitory processes associated with interference and negative priming effects. As reviewed in Chapter 3, Roberts and Pennington (1996) define inhibition as the suppression of prepotent alternative responses, which occurs by default when working memory processes are appropriately activated and maintained. In this framework, response selection results from an interaction among an individual's working memory resources, the working memory demands for producing correct responses, and the strength ofthe competing prepotencies. The direct implication of this model is that increasing the working memory demands of a task will compromise working memory performance and hence,

decrease the ability to inhibit competing responses. A similar prediction is made from the controlled attention model of Engle et al. (1999b). According to this model, individual differences in working memory capacity reflect differences in the capacity

for controlled attention in the face of interference or distraction and hence, corresp to the ability to ignore irrelevant information. Thus, increasing the working memory

demands of a task will place a greater demand on limited attentional resources, leavin less available for the inhibition of irrelevant information. Support for these predictions comes from the studies of Roberts et al. (1994), who demonstrated that increasing the load on working memory impaired the ability to resist reflexive saccades, and Conway et al. (1999) who showed that increasing the working memory load eliminated negative priming. However, both studies manipulated working memory load by introducing a secondary task. As a result, the reduced ability 91

to inhibit competing responses m a y be due to the increase in load on domain-free working memory resources as Conway et al. (1999) would predict, or it may be that attention was re-allocated from one task to the other and it was this task-switching

requirement that led to inefficient inhibitory processing. Task switching is known to b an effortful process as shown by substantial task switch costs even under predictable switching conditions (Allport et al., 1994; Rogers & Monsell, 1995), and has been shown to interfere with executive functioning (Baddeley, 1996). Thus, the reduced

ability to inhibit irrelevant stimuli demonstrated in these studies might be a direct r ofthe task-switching requirement. The purpose ofthe present study was to achieve a more general conclusion concerning the influence of working memory load on the inhibitory processing of irrelevant information, by manipulating working memory load within the same task. In Experiment 1, a modified version ofthe Stroop task was developed to incorporate a manipulation of memory load. According to Roberts and Pennington (1996) tasks such as the Stroop and Antisaccade tasks have strong prepotent responses but relatively small working memory demands and as such, will demonstrate impaired responding with even a slight deficiency in working memory. The traditional Stroop

task involves presenting participants with lists of colour-words presented in various i

colours (Stroop, 1935). The participant is typically required to respond to one dimensio

ofthe stimulus (i.e. the ink-colour), while ignoring the other dimension (i.e. the colo

word). In comparison to a neutral baseline, this provides a measure of interference tha reflects the degree of response competition arising from the processing ofthe ignored word. A measure of negative priming can also be obtained in this task by including a condition in which the ignored colour-word becomes the target ink-colour on the subsequent trial. According to active inhibition accounts of negative priming, the 92

slowing of responses to targets that were previously ignored distractors provides an index ofthe inhibition applied to the distracting information on the previous trial (Houghton & Tipper, 1994; Tipper, 1985). This will provide a comparable measure to that used by Engle et al. (1995) and Conway et al. (1999) and thus allow close comparisons across the different experiments. The concept of load implies more than just increased difficulty; it implies that the system must carry out additional operations or apply operations to more stimuli (Lavie, 1995). In line with this view, the working memory demand involved in the Stroop task was manipulated by incorporating a memory load without creating a secondary task distinct from the primary Stroop task. This was achieved by having

participants remember the colours ofthe Stoop stimuli that they named during a trial of

successive Stroop items, and then recall the colours in serial order at the end of each

trial. In this section ofthe task, increasing list lengths were used to assess the eff

increasing memory load on inhibition, as it was unclear which list length would provide an appropriately difficult task and place demands on the working memory system. According to Roberts and Pennington (1996), increasing the working memory demands ofthe task should reduce the level of moment-to-moment activation thought to be

involved in the Stroop task, resulting in less inhibition ofthe irrelevant colour word more naming errors. Similarly, in line with the proposal of Engle et al. (1999a), the introduction of a memory load should draw on general attentional resources, leaving less available for the inhibition ofthe irrelevant colour word. This would lead to increased interference from the irrelevant colour word during selection and reduced negative priming. In terms of memory performance, the interference and negative priming conditions are likely to place heavier demands on working memory than the

93

control condition. If working m e m o r y and inhibition are interactive this should lead to poorer recall in these conditions and an interaction with increasing list length.

4.2 Method 4.2.1 Participants

The participants were 50 undergraduate students at the University of Wollongong w h o received course credit for participation. O f these participants, 14 were subsequently excluded from the analysis on the basis that they became aware ofthe relationship between consecutive trials in the negative priming condition. Awareness of the relationship makes it possible to predict the colour on the following trial and reduce the naming latencies in this condition. Awareness was determined by post-test questioning in which participants were asked if they noticed any patterns or sequences in the trials. Participants were excluded if they could describe the negative priming pattern. All participants participating in the study had normal or corrected to normal vision and spoke English as their first language.

4.2.2 Materials

Thirty Stroop stimuli were constructed using the words red, yellow, green, blue, black and purple, written in each ofthe named colours except for that congruent to the written word. There were three conditions: a control condition, an interference condition and a negative priming condition. In the interference condition, trials were organised so that there were no colour-to-word relationships between consecutive items within each trial. In the negative priming condition, trials were organised so that the distractor colour word became the target ink colour ofthe following item. N o colours or words were repeated within a trial and the number of colour words, ink colours and 94

consecutive colour-to-colour relationships were approximately equal across conditions and list lengths. A control condition was also included in which each stimulus was a row of X's, three to six characters long, in one ofthe six colours. The experiment was divided into two sections, the Stroop task with and without a memory load. The Stroop task without a memory load consisted of one block of 30

trials with 10 trials in each ofthe control, interference and negative priming conditio Each trial was five items in length. The order of trial presentation was randomised for each participant. In the memory load section, the same three conditions were used however the length ofthe trials within each condition was varied. List lengths of four, five and six items were used with 10 trials of each list length occurring in each condition. Trials were assigned to one of three blocks so that equal numbers of trials from each list length and condition appeared in each block. The order of trials within each block was randomised for each participant and the presentation order of blocks was counterbalanced between participants. Although most reaction time experiments include at least 15 trials per condition

to ensure stable reaction times, it was thought that this would be particularly arduous participants in this experiment, as it would mean that they would be required to complete 45 Stroop trials without a memory load and 135 memory trials varying from 4- to 6-items in length in the memory load condition. Pilot testing revealed that this

a difficult task for participants and so, to avoid fatigue and stress, only 10 trials we included in each condition. However, to compensate for the reduced number of trials

per condition, a large sample size was employed. In addition, it should be noted that th reduced number of trials per condition would only affect within-list analyses across serial position. Because four or more response times were recorded on each trial, all

95

analyses ofthe main effects of condition and/or task are based on averages of at least 40 responses, which is sufficient to ensure the stability of these response times. The experiment was programmed using SuperLab Experimental Laboratory Software and stimuli were presented on a 14 inch flat square high contrast colour monitor controlled by a Macintosh Quadra 650. The stimuli were displayed in uppercase 18-point Palatino font and subtended a horizontal visual angle of approximately 1.5° for the shortest stimulus (RED) to 3.2° for the longest stimuli (YELLOW) at a viewing distance of 50cm. A microphone headset, calibrated within the SuperLab program, was used to measure verbal naming latencies.

4.2.3 Procedure

Participants were tested individually in a single session of approximately 40 minutes. At the beginning ofthe session, each participant was presented with a test

screen consisting of six rows of X's in the target ink colours. Participants were asked name each colour as a test of colour vision and were excluded if they failed to name each colour correctly. Participants were instructed that there were two tasks in the experiment, both of which required them to name the ink colour of each stimulus as quickly and as accurately as possible while keeping other vocal noises to a minimum. They were further instructed that in the second task, they would also be required to

recall the colours they had named, in serial order at the end of each trial. If they we

unable to recall an item, they were instructed to say 'blank'. Participants completed t single block of Stroop trials first followed by the three blocks of Stroop trials with memory load. Practice trials of each condition and each list length were given at the beginning ofthe experiment.

96

Each trial began with a fixation cross in the centre ofthe screen followed by the first item. Items remained on the screen until the participant made a response, after which there was a 500msec inter-stimulus interval (ISI) followed by the presentation the next item. At the end ofthe trial a row of question marks were presented. In the Stroop task without the memory load, the participants pressed the space bar and

continued with the next trial. In the memory load trials, the participants recalled th colours from the trial in serial order before proceeding to the next trial. The experimenter recorded the participants' responses as well as any naming errors and

voice-key errors triggered by extraneous vocal activity during presentation. Trials w naming error or voice-key error were excluded from the analyses. Naming latency was recorded by the computer and was measured from the presentation ofthe stimulus until the oral response activated the speech trigger.

4.3 Results

The assumption of sphericity was examined for every analysis involving a repeated measures variable with three or more levels by way of Mauchly's test of

sphericity. If the sphericity assumption was violated, a Greenhouse-Geisser correctio was evaluated. Where this correction resulted in a different outcome, the corrected

degrees of freedom and resulting probability level are reported. Unless otherwise sta the alpha level for all statistical tests was .05.

4.3.1 Naming Latency

The dependent variable of primary interest was naming latency. Median reaction

times were computed for each participant as a function of memory load (zero load, 4, 5

or 6 item lists), condition (control, interference, negative priming) and serial posit 97

Response latencies longer than 2 seconds were excluded from the analysis. In the initial analysis, reaction times were averaged across serial position to examine the effect

both introducing a memory load and increasing the memory load across list length. Th

means and standard errors of each list length within each condition are shown in Fig 4.1 for the Stroop tasks performed with and without a memory load.

Stroop Task — A — Control — O — Interference —•— Negative Priming Stroop Task + m e m o r y load —±- Control —•— Interference • Negative Priming

4.0

5.0

6.0

List Length

Figure 4.1: Naming latency in the Stroop task with and without a memory load as a function of condition and list length, collapsed across serial position.

The effects of introducing a memory load and increasing a memory load on naming latencies were examined in two separate analyses. To examine the effect of introducing a memory load, a two-way repeated measures analysis of variance

(AVOVA) was conducted on data from the 5-item lists, with task (Stroop alone, Stroop

+ memory load), and condition (control, interference, negative priming) as the variab

A main effect of task was found [F(l,35) = 46.06, p

3

4

Serial Position

Figure 5.37: M e a n proportion of colours correctly recalled in the unmasked colour recall task as a function of serial position and condition.

1.0 - A - Control -•— Interference -•— Negative Priming

0.9

O

0.8

O

o

O 0.7 '•E

o Q.

2

0.6

Q. 0.5 -

0.4

Serial Position Figure 5.38: M e a n proportion of colours correctly recalled in the masked colour recall task as a function of serial position and condition.

190

A three-way repeated measures A N O V A was performed with masking (masked,

unmasked), condition (control, interference, negative priming), and serial position as th variables. There was no effect of masking (p > .10), which suggests that masking the

stimuli after presentation did not affect the overall level of recall accuracy. However, there was a marginally significant interaction between masking and serial position [F(3,72) = 2.69, p = .053], which suggests that masking the stimuli after presentation does have some effect on the serial position curves in the two masking conditions. This

appears to be due to poorer recall of items at serial position two in the masked conditi [F(l,24) = 5.61, p~^ ^ ^ \ •

1

Q.

T

T


\

—A

i — "

7

2 Q.

0.4

n9

2

3

4

Serial Position

Figure 5.39: Recall accuracy of low-span participants in the unmasked recall task as a function of condition and serial position.

Figure 5.40: Recall accuracy of low-span participants in the masked recall task as a function of condition and serial position.

192

1.0

0.8

u

£ o o

0.6

c o

'€ o

0.4 - A — Control -•— Interference -•— Negative Priming

Q.

o 0.2

Serial Position Figure 5.41: Recall accuracy of high-span participants in the unmasked recall task as a function of condition and serial position.

1.0

0.8

o

£ L.

o o c o t o

0.6

0.4 Control Interference Negative Priming

Q.

O

0.2

Serial Position Figure 5.42: Recall accuracy of high-span participants in the masked recall task as a function of condition and serial position.

193

A four-way mixed design A N O V A was conducted with span group as the between-group variable. The main effect of span group was significant [F(l,17) = 4.82, p o 0) 4 Q.

J rV

*

re *J

c CD

2 CD

a.

Task Figure 5.44: Naming errors in the naming and colour recall tasks as a function of condition and span group.

The main effect of span group was not significant (p > .10), however, there was

a significant interaction between task and span group [F(l,17) = 5.27, p

4

CD

C/3

Load

Figure 6.6: M e a n d' values for high-span participants in the single letter and letter triplet conditions ofthe n-back task as a function of working memory load.

A three-way mixed design A N O V A was conducted with interference condition and working memory load as the within-subjects variables and span group as the between-group variable. Consistent with the main analysis, the main effect of

interference condition was significant [F(l,29) =18.31, p

2 70 5 175 10 350 647

MS 0.33 4.99 7.62 6.81 0.20 5.96

F 0.16 0.07 1.52 0.04 0.02 0.02

2.30 37.20 1.17

89.6Ci

Note: Cond = Condition, Serpos = Serial Position Table A 10: Summary table for repeated measures A N O V A on naming errors by task and condition in Experiment 1 Source Within Subjects Task Error (Task) Cond Error (Cond) Task x Cond Error (Task x Cond) Total

df

SS

26

0.17

1 26 2 52 2 52 161

MS 0.03 0.20 0.19 0.33 0.02 0.46

F 0.03 0.01 0.10 0.01 0.01 0.01

3.34 15.33 1.08

1.40

Note: Cond = Condition Table All: Summary table for repeated measures ANOVA on naming errors by condition and list length in Experiment 1 Source Within Subjects Cond Error (Cond) List Error (List) Cond x List Error (Cond x List) Total

df

SS

26

0.04

MS 0.27 0.30 0.02 0.28 0.03 0.72

2 52 2 52 4

104 242

1.66

Note: Cond = Condition, List = List Length

310

F 0.14 0.01 0.01 0.01 0.01 0.01

23.03 2.16 0.96

Table A 12: Summary table for repeated measures A N O V A on naming latencies by

task and condition in Experiment 1 for aware participants (Stroop task with an a memory load) Source Within Subjects Task Error (Task) Cond Error (Cond) Task x Cond Error (Task x Cond) Total

df

SS

13

1669084.05 37990.77 49953.51 273972.04 154928.95 1873.98 17327.11 2205130.41

1 13 2 26 2 26 83

MS 37990.77 3842.58 136986.02 5958.81 936.99 666.43

F 9.89 22.99 1.41

Note: Cond = Condition Table A 13: Summary table for repeated measures A N O V A on naming latencies by condition and list length in Experiment 1 for aware participants Source Within Subjects Cond Error (Cond) List Error (List) Cond x List Error (Cond x List) Total

df

SS

13

2979227.37 334548.74 248100.11 14457.55 47365.00 1664.59 50511.56 3675874.92

2 26 2 26 4

52 125

Note: Cond = Condition, List = List Length

311

MS 167274.37 9542.31 7228.78 1821.73 416.15 971.38

F 17.53 3.97 0.43

Appendix B Major statistical analyses performed in Experiment 2a, 2b, 2c and 2d

Table B 1: Summary table for repeated measures A N O V A on naming latencies by task, condition and serial position in Experiment 2a Source df SS 4182864.12 Within Subjects 39 3824238.83 Task 1 1291732.35 39 Error (Task) 598315.79 2 Cond 225573.56 78 Error (Cond) 902341.23 3 Serpos 983680.96 117 Error (Serpos) 17726.25 2 Task x Cond 165983.94 78 Error (Task x Cond) 67579.59 3 Task x Serpos 612005.31 117 Error (Task x Serpos) 10356.06 6 Cond x Serpos 339766.00 234 Error (Cond x Serpos) 3520.03 6 Task x Cond x Serpos 359228.70 234 Error (Task x Cond x Serpos) 13584912.72 959 Total Note: Cond = Condition, Serpos = Serial Position

MS

F

3824238.83 33121.34 299157.90 2891.97 300780.41 8407.53 8863.13 2128.00 22526.53 5230.82 1726.01 1451.99 586.67 1535.17

115.46 103.44 35.78 4.17 4.31 1.19 0.38

Table B 2: Summary table for repeated measures A N O V A on recall accuracy by condition and serial position in Experiment 2a Source Within Subjects Cond Error (Cond) Serpos Error (Serpos) Cond x Serpos Error (Cond x Serpos) Total

df

SS

39

28.97 0.49 3.16 1.22 5.82 0.09 4.39

2 78 3 117 6 234

479

44.14

Note: Cond = Condition, Serpos = Serial Position

312

F

MS 0.24 0.04 0.41 0.05 0.01 0.02

5.99 8.20 0.77

Table B 3: Summary table for repeated measures A N O V A on naming errors by task and condition in Experiment 2a Source Within Subjects

Task Error (Task) Cond Error (Cond) Task x Cond Error (Task x Cond) Total

df 39 1 39 2 78 2 78 239

SS

MS

F

0.08 0.01 0.12 0.04 0.25 0.01 0.28

0.01 0.00 0.02 0.00 0.00 0.00

4.20 6.79 1.15

0.79

Note: Cond = Condition

Table B 4: Summary table for repeated measures ANOVA on naming latencies by tas condition and serial position in Experiment 2a for aware participants

df SS Source 3631608.23 31 Within Subjects 1489885.55 Task 1 636096.17 31 Error (Task) 555480.38 2 Cond 212244.58 62 Error (Cond) 687937.48 3 Serpos 561708.27 93 Error (Serpos) 13353.26 2 Task x Cond 122831.86 62 Error (Task x Cond) 50368.37 3 Task x Serpos 220894.70 93 Error (Task x Serpos) 3464.16 6 Cond x Serpos 198843.39 186 Error (Cond x Serpos) 10861.45 6 Task x Cond x Serpos 183443.76 186 Error (Task x Cond x Serpos) 8579021.61 767 Total Note: Cond = Condition, Serpos =- Serial Position

313

MS 1489885.55 20519.23 277740.19 3423.30 229312.49 6039.87 6676.63 1981.16 16789.46 2375.21 577.36 1069.05 1810.24 986.26

F 72.61 81.13 37.97 3.37 7.07 0.54 1.84

Table B 5: Summary table for repeated measures A N O V A on naming latencies by task, condition and serial position in Experiment 2b Source df SS Within Subjects 39 41199462.98 Task 2 4012320.54 Error (Task) 78 3409600.39 Cond 2 623335.01 Error (Cond) 78 242009.21 Serpos 3 447292.99 Error (Serpos) 117 678452.53 4 1963.96 Task x Cond Error (Task x Cond) 156 269739.40 6 97353.10 Task x Serpos 234 635841.81 Error (Task x Serpos) 6 16635.14 Cond x Serpos 234 162661.97 Error (Cond x Serpos) 12 9137.98 Task x Cond x Serpos 468 360634.49 Error (Task x Cond x Serpos) 1439 52166441.50 Total Note: Cond = Condition, Serpos = Serial Position

314

MS 2006160.27 43712.83 311667.51 3102.68 149097.66 5798.74 490.99 1729.10 16225.52 2717.27 2772.52 695.14 761.50 770.59

F 45.89 100.45 25.71 0.28 5.97 3.99 0.99

Table B 6: Summary table for mixed design A N O V A on naming latencies by span group, task, condition and serial position in Experiment 2b Source df SS MS Between Subjects 28 3489199.30 Spangrp 1 47776.55 47776.55 Error (Spangrp) 27 3441422.75 127460.10 Within Subjects 1015 7418775.12 Task 2 2749637.46 1374818.73 Task x Spangrp 2 74122.89 37061.44 Error (Task) 54 2368625.04 43863.43 Cond 2 426975.41 213487.71 Cond x Spangrp 2 23207.21 11603.60 Error (Cond) 54 136889.87 2535.00 Serpos 3 292287.11 97429.04 Serpos x Spangrp 3 26054.55 8684.85 Error (Serpos) 81 359385.47 4436.86 Task x Cond 4 4974.08 1243.52 Task x Cond x Spangrp 4 5557.00 1389.25 Error (Task x Cond) 108 123732.40 1145.67 Task x Serpos 6 72002.60 12000.43 Task x Serpos x Spangrp 6 21221.79 3536.97 Error (Task x Serpos) 162 331471.89 2046.12 Cond x Serpos 6 10182.31 1697.05 Cond x Serpos x Spangrp 6 4028.10 671.35 162 Error (Cond x Serpos) 114906.47 709.30 12 9596.59 799.72 Task x Cond x Serpos 12 Task x Cond x Serpos x Spangrp 7230.00 602.50 324 256686.88 792.24 Error (Task x Cond x Serpos) 1043 10907974.42 Total Note: Spangrp = Span Group, Cond = Condition, Serpos = Serial Position

315

F 0.38

31.34 0.85 84.22 4.58 21.96 1.96 1.09 1.21 5.87 1.73 2.39 0.95 1.01 0.76

Table B 7: Summary table for repeated measures A N O V A on recall accuracy by task, condition and serial position in Experiment 2b Source Within Subjects Task Error (Task) Cond Error (Cond) Serpos Error (Serpos) Task x Cond Error (Task x Cond) Task x Serpos Error (Task x Serpos) Cond x Serpos Error (Cond x Serpos) Task x Cond x Serpos Error (Task x Cond x Serpos) Total

df

SS

39

14.16 18.58 3.057 0.18 2.64 2.16 1.85 0.03 2.68 0.56 1.51 0.04 2.43 0.04 2.33 52.25

1 39 2 78 3 117 2 78 3 117 6 234 6 234 959

Note: Cond = Condition, Serpos = Serial Position

316

MS

F 18.58 0.08 0.09 0.03 0.72 0.02 0.01 0.03 0.19 0.01 0.01 0.01 0.01 0.01

236.95 2.63 45.55 0.39 14.30 0.64 0.71

Table B 8: Summary table for mixed design A N O V A on recall accuracy by span group, task, condition and serial position in Experiment 2b Source df SS MS F Between Subjects 28 11.27 Spangrp 1 0.58 0.58 27 10.69 0.40 Error (Spangrp) Within Subjects 667 26.22 Task 1 12.91 12.91 Task x Spangrp 1 0.00 0.00 27 1.87 0.07 Error (Task) Cond 2 0.07 0.03 Cond x Spangrp 2 0.04 0.02 54 1.78 0.03 Error (Cond) Serpos 3 1.45 0.48 Serpos x Spangrp 3 0.02 0.01 81 1.37 0.02 Error (Serpos) Task x Cond 2 0.04 0.02 Task x Cond x Spangrp 2 0.01 0.01 54 1.92 0.04 Error (Task x Cond) Task x Serpos 3 0.26 0.09 Task x Serpos x Spangrp 3 0.05 0.02 81 0.94 0.01 Error (Task x Serpos) Cond x Serpos 6 0.01 0.00 Cond x Serpos x Spangrp 6 0.06 0.01 162 1.66 0.01 Error (Cond x Serpos) Task x Cond x Serpos 6 0.01 0.00 Task x Cond x Serpos x Spangrp 6 0.12 0.02 162 L63 0M Error (Task x Cond x Serpos) Note: Spangrp = Span Group, Cond = Condition, Serpos = Serial Position Total 695 37.49

317

1.47

186.62 0.00 1.04 0.56 28.62 0.32 0.53 0.16 7.44 1.43 0.24 1.04 0.19 1.91

Table B 9: Summary table for repeated measures A N O V A on naming errors by task and condition in Experiment 2b Source Within Subjects Task Error (Task) Cond Error (Cond) Task x Cond Error (Task x Cond) Total

df

SS

39

0.77

2 78 2 78 4 156 359

MS 0.02 0.21 0.15 0.24 0.01 0.58

F 0.01 0.00 0.08 0.00 0.00 0.00

4.21 25.20 0.49

1.98

Note: Cond = Condition Table B10: Summary table for mixed design ANOVA on naming errors by span group, task and condition in Experiment 2b Source Between Subjects Spangrp Error (Spangrp) Within Subjects Task Task x Spangrp Error (Task) Cond Cond x Spangrp Error (Cond) Task x Cond Task x Cond x Spangrp Error (Task x Cond) Total

df 28

1 27 232 2 2 54 2 2 54 4 4 108 260

Note: Soanero = Scan Group, C',ond = Condition

318

SS 0.22 0.05 0.17 0.98 0.03 0.01 0.17 0.10 0.01 0.17 0.01 0.05 0.43 1.20

MS

F 0.05 0.01

l.AA

0.02 0.00 0.00 0.05 0.00 0.00 0.00 0.01 0.00

5.15 0.99 16.49 1.31 0.48 3.09

Table B 11: Summary table for repeated measures A N O V A on naming latencies by task, condition and serial position in Experiment 2b for aware participants Source Within Subjects

df 27

SS

36641141.40 Task 2 3831018.13 Error (Task) 54 2745742.54 Cond 2 738150.16 Error (Cond) 54 182922.01 Serpos 3 704583.68 Error (Serpos) 81 691056.31 Task x Cond 4 5997.24 Error (Task x Cond) 108 283130.01 Task x Serpos 6 46075.41 Error (Task x Serpos) 162 457684.15 Cond x Serpos 6 2608.05 Error (Cond x Serpos) 162 176119.51 12 Task x Cond x Serpos 17573.98 324 Error (Task x Cond x Serpos) 353102.05 46876904.63 1007 Total = Note: Cond = Condition, Serpos = Serial Position

MS 1915509.07 50847.08 369075.08 3387.45 234861.23 8531.56 1499.31 2621.57 7679.24 2825.21 434.67 1087.16 1464.50 1089.82

F 37.67 108.95 27.53 0.57 2.72 0.40 1.34

Table B 12: Summary table for repeated measures A N O V A on naming latencies by task, condition and serial position in Experiment 2c

SS df Source 23316878.99 39 Within Subjects 852914.39 2 Task 344989.33 78 Error (Task) 633459.24 2 Cond 151128.69 78 Error (Cond) 1671721.67 3 Serpos 902931.95 117 Error (Serpos) 3873.61 4 Task x Cond 159171.71 156 Error (Task x Cond) 338526.94 6 Task x Serpos 402344.00 234 Error (Task x Serpos) 2820.51 6 Cond x Serpos 173300.89 234 Error (Cond x Serpos) 13373.26 12 Task x Cond x Serpos 315105.59 468 Error (Task x Cond x Serpos) 29282540.77 1439 Total : Note: Cond = Condition, Serpos = Serial Position

319

MS 426457.20 4422.94 316729.62 1937.55 557240.56 7717.37 968.40 1020.33 56421.16 1719.42 470.09 740.60 1114.44 673.30

F 96.42 163.47 72.21 0.95 32.81 0.64 1.66

Table B 13: Summary table for mixed design A N O V A on naming latencies by span group, task, condition and serial position in Experiment 2c Source df SS MS Between Subjects 34 19903132.31 Spangrp 1 46087.31 46087.31 Error (Spangrp) 33 19857045.00 601728.64 Within Subjects 1225 5540024.42 Task 2 805951.67 402975.84 Task x Spangrp 2 19371.50 9685.75 Error (Task) 66 300457.62 4552.39 Cond 2 509948.23 254974.11 Cond x Spangrp 2 1355.37 677.69 Error (Cond) 66 124724.41 1889.76 Serpos 3 1609766.53 536588.84 Serpos x Spangrp 3 11223.56 3741.19 Error (Serpos) 99 840728.78 8492.21 Task x Cond 4 8122.14 2030.53 Task x COnd x Spangrp 4 3651.30 912.83 Error (Task x Cond) 132 131182.73 993.81 Task x Serpos 6 329683.90 54947.32 Task x Serpos x Spangrp 6 6485.08 1080.85 Error (Task x Serpos) 198 375195.74 1894.93 Cond x Serpos 6 3251.20 541.87 Cond x Serpos x Spangrp 6 6518.94 1086.49 Error (Cond x Serpos) 799.10 198 158220.93 1005.76 Task x Cond x Serpos 12069.06 12 12 314.99 3779.88 Task x Cond x Serpos x 702.87 396 278335.85 Error (Task x Cond x Serpos) 25443156.73 1259 Total Note: Spangrp = Span Group, Cond = Condition, Serpos = Serial Position

320

F 0.08

88.52 2.13 134.92 0.36 63.19 0.44 2.04 0.92 29.00 0.57 0.68 1.36 1.43 0.45

Table B 14: Summary table for repeated measures A N O V A on recall accuracy by condition in Experiment 2c (1-item preload task) Source df SS Within Subjects 39 0.08 Cond 2 0.01 Error (Cond) 78 0.31 Total 119 0.40 Note: Cond = Condition, Serpos = Serial Position

MS

F 0.01 0.00

1.56

Table B 15: Summary table for repeated measures A N O V A on recall accuracy by condition in Experiment 2c (4-item preload task) Source Within Subjects Cond Error (Cond) Serpos Error (Serpos) Cond x Serpos Error (Cond x Serpos) Total

df 39

SS Q.lt 0.43 3.42 1.73 1.85 0.10 2.96 11.25

2 78 3 117 6 234 479

MS

t

F 0.22 0.04 0.58 0.02 0.02 0.01

4.94 36.43 1.28

Note: Cond = Condition, Serpos = Serial Position Table B 16: Summary table for mixed design ANOVA on recall accuracy by span group and condition in Experiment 2c (1-item preload task) Source df SS MS F Between Subjects Spangrp Error (Spangrp) Within Subjects Cond Cond x Spangrp Error (Cond) Total

34

1 33 70 2 2 66 104

Note: Spangrp = Span Group, Cond = Condition

321

0.08 0.02 0.06 0.31 0.02 0.00 0.29 0.39

0.02 0.00

8.40

0.01 0.00 0.00

2.04 0.43

Table B 17: Summary table for mixed design A N O V A on recall accuracy by span group and condition in Experiment 2c (4-item preload task) Source df Between Subjects 34 Spangrp 1 Error (Spangrp) 33 Within Subjects 385 Cond 2 Cond x Spangrp 2 Error (Cond) 66 Serpos 3 Serpos x Spangrp 3 Error (Serpos) 99 6 Cond x Serpos 6 Cond x Serpos x Spangrp Error (Cond x Serpos) 198 419 Total Note: Spangrp = Span Group, Cond = Condition,

SS

MS

0.69 0.07 0.07 0.62 0.02 9.28 0.35 0.18 0.22 0.11 0.04 2.75 0.50 1.50 0.01 0.04 0.02 1.73 0.02 0.11 0.01 0.04 0.01 2.54 9.97 Serpos = Serial Position

F 3.94

4.23 2.63 28.63 0.69 1.46 0.51

Table B 18: Summary table for repeated measures ANOVA on naming errors by task and condition in Experiment 2c Source Within Subjects Task Error (Task) Cond Error (Cond) Task x Cond Error (Task x Cond) Total

df 39

2 78 2 78 4 156 359

0.01 0.22 0.09 0.25 0.00 0.45 1.04

Note: Cond = Condition

322

F

MS

SS 0.02

0.00 0.00 0.04 0.00 0.00 0.00

1.34 13.86 0.10

Table B 19: Summary table for mixed design A N O V A on naming errors by span group, task and condition in Experiment 2c Source Between Subjects Spangrp Error (Spangrp) Within Subjects Task Task x Spangrp Error (Task) Cond Cond x Spangrp Error (Cond) Task x Cond Task x Cond x Spangrp Error (Task x Cond) Total

df 34

SS

MS

0.02 0.00 0.02 0.90 0.01 0.00 0.20 0.07 0.01 0.22 0.00 0.00 0.39 0.92

1 33 280 2 2 66 2 2 66 4 4 132 314

F 0.00 0.00

3.14

0.00 0.00 0.00 0.04 0.01 0.00 0.00 0.00 0.00

1.18 0.54 10.97 1.72 0.37 0.09

Note: Spangrp = Span Group, Cond = Condition Table B 20: Summary table for repeated measures ANOVA on naming latencies by task, condition and serial position in Experiment 2c for aware participants Source Within Subjects Task Error (Task) Cond Error (Cond) Serpos Error (Serpos) Task x Cond Error (Task x Cond) Task x Serpos Error (Task x Serpos) Cond x Serpos Error (Cond x Serpos) Task x Cond x Serpos Error (Task x Cond x Serpos) Total

df 11

2 22 2 22 3 33 4 44 6 66 6 66 12 132 431

SS 66723.80 669876.05 1358449.34 212084.52 243010.12 326902.26 310925.24 9421.98 59941.98 13536.70 210441.47 17482.28 210439.31 8090.06 141596.94 3858922.05

Note: Cond = Condition, Serpos = Serial Position

323

MS 334938.03 61747.70 106042.26 11045.92 108967.42 9421.98 583.03 1362.32 2256.12 3188.51 2913.71 3188.47 674.17 1072.70

F 5.42 9.60 11.57 0.43 0.71 0.91 0.63

Table B 21: Summary table for repeated measures A N O V A on naming latencies by mask, task, condition and serial position in Experiment 2d Source Within Subjects Mask Error (Mask) Task Error (Task) Cond Error (Cond) Serpos Error (Serpos) Mask x Task Error (Mask x Task) Mask x Cond Error (Mask x Cond) Task x Cond Error (Task x Cond) Mask x Task x Cond Error (Mask x Task x Cond) Mask x Serpos Error (Mask x Serpos) Task x Serpos Error (Task x Serpos) Mask x Task x Serpos Error (Mask x Task x Serpos) Cond x Serpos Error (Cond x Serpos) Mask x Cond x Serpos Error (Mask x Cond x Serpos) Task x Cond x Serpos Error (Task x Cond x Serpos) Mask x Task x Cond x Serpos Error (Mask x Task x Cond x Serpos) Total

df 24 1 24 1 24 2 48 3 72 1 24 2 48 2 48 2 48 3 72 3 72 3 72 6 144 6 144 6 144 6 144 1199

Note: Cond = Condition, Serpos = Serial Position

324

SS

MS

F

553561.17 975612.21 724728.60 2622581.50 2979256.44 421983.83 168195.45 461599.77 1055989.71 7650.75 590465.35 872.23 63780.55 5049.22 151199.88 5058.49 80322.31 7708.86 257577.50 43403.51 738615.22 6279.61 183552.78 30028.12 233700.93 5930.01 191573.54 15570.39 239441.35 8587.96 234292.24 13064169.48

975612.21 30197.03 2622581.50 124135.69 210991.92 3504.07 153866.59 14666.52 7650.75 24602.72 436.12 1328.76 2524.61 3150.00 2529.24 1673.38 2569.62 3577.47 14467.84 10258.55 2093.20 2549.34 5004.69 1622.92 988.34 1330.37 2595.07 1662.79 1431.33 1627.03

32.31 21.13 60.21 10.49 0.31 0.33 0.80 1.51 0.72 1.41 0.82 3.08 0.74 1.56 0.88

Table B 22: Summary table for mixed design A N O V A on naming latencies by span group, mask, task, condition and serial position in Experiment 2d Source Between Subjects Spangrp Error (Spangrp) Within Subjects Mask Mask x Spangrp Error (Mask) Task Task x Spangrp Error (Task) Cond Cond x Spangrp Error (Cond) Serpos Serpos x Spangrp Error (Serpos) Mask x Task Mask x Task x Spangrp Error (Mask x Task) Mask x Cond Mask x Cond x Spangrp Error (Mask x Cond) Task x Cond Task x Cond x Spangrp Error (Task x Cond) Mask x Task x Cond Mask x Task x Cond x Spangrp Error (Mask x Task x Cond) Mask x Serpos Mask x Serpos x Spangrp Error (Mask x Serpos) Task x Serpos Task x Serpos x Spangrp Error (Task x Serpos) Mask x Task x Serpos Mask x Task x Serpos x Spangrp Error (Mask x Task x Serpos) Cond x Serpos Cond x Serpos x Spangrp Error (Cond x Serpos) Mask x Cond x Serpos Mask x Cond x Serpos x Spangrp Error (Mask x Cond x Serpos) Task x Cond x Serpos

df

SS

18

1 17 893 1 1 17 1 1 17 2 2 34 3 3 51 1 1 17 2 2 34 2 2 34 2 2 34 3 3 51 3 3 51 3 3 51 6 6 102 6

6 102 6

325

447583.29 11351.14 436232.15 9837350.41 681980.48 25393.93 620207.75 1896613.58 37440.24 2293852.48 289922.95 12750.45 127656.64 302802.24 121259.07 764022.13 4062.01 50339.35 495928.23 1696.76 257.41 54730.64 5385.44 7595.24 108446.54 5721.73 61.11 75271.69 18653.74 26836.23 188559.40 37385.30 71454.76 578984.87 11459.60 25163.33 132895.05 23182.43 4303.61 184266.11 6228.06 8673.23 133322.98 12639.50

MS

F

11351.14 25660.72

0.44

681980.48 25393.93 36482.81 1896613.5 37440.24 134932.50 144961.48 6375.23 3754.61 100934.08 40419.69 14980.83 4062.01 50339.35 29172.25 848.38 128.71 1609.73 2692.72 3797.62 3189.60 2860.87 30.55 2213.87 6217.91 8945.41 3697.24 12461.77 23818.25 11352.64 3819.87 8387.78 2605.79 3863.74 717.27

18.69 0.70

1806.53 1038.01 1445.54 1307.09 2106.58

14.06 0.28 38.61 1.70 6.74 2.70 0.14 1.73 0.53 0.08 0.84 1.19 1.29 0.01 1.68 2.42 1.10 2.10 1.47 3.22 2.14 0.40 0.79 1.11 1.21

Task x Cond x Serpos x Spangrp 6 15246.50 Error (Task x Cond x Serpos) 102 177298.81 Mask x Task x Cond x Serpos 6 12633.88 Mask x Task x Cond x Serpos x 6 6779.82 Error (Mask x Task x Cond x 102 177985.11 Total 911 10284933.70 Note: Spangrp = Span Group, Cond = Condition, Serpos = Serial

2541.08 1738.22 2105.65 1129.97 1744.95

1.46 1.21 0.65

Position

Table B 23: Summary table for repeated measures ANOVA on recall accuracy by mask, condition and serial position in Experiment 2d Source Within Subjects Mask Error (Mask) Cond Error (Cond) Serpos Error (Serpos) Mask x Cond Error (Mask x Cond) Mask x Serpos Error (Mask x Serpos) Cond x Serpos Error (Cond x Serpos) Mask x Cond x Serpos Error (Mask x Cond x Serpos) Total

df

SS

24

1.25

1 24 2 48 3 72 2 48 3 72 6 144 6 144 599

0.11 2.31 0.23 2.20 1.89 3.78 0.01 2.00 0.15 1.30 0.13 2.83 0.15 3.86 22.20

Note: Cond = Condition, Serpos = Serial Position

326

MS

F 0.11 0.10 0.11 0.05 0.63 0.05 0.00 0.04 0.05 0.02 0.02 0.02 0.02 0.03

1.10 2.49 11.99 0.07 2.69 1.10 0.90

Table B 24: Summary table for mixed design A N O V A on recall accuracy by span group, mask, condition and serial position in Experiment 2d Source df Between Subjects 18 Spangrp 1 Error (Spangrp) 17 Within Subjects 437 Mask 1 Mask x Spangrp 1 Error (Mask) 17 Cond 2 Cond x Spangrp 2 Error (Cond) 34 Serpos 3 Serpos x Spangrp 3 Error (Serpos) 51 Mask x Cond 2 Mask x Cond x Spangrp 2 Error (Mask x Cond) 34 Mask x Serpos 3 Mask x Serpos x Spangrp 3 Error (Mask x Serpos) 51 Cond x Serpos 6 Cond x Serpos x Spangrp 6 Error (Cond x Serpos) 102 Mask x Cond x Serpos 6 Mask x Cond x Serpos x Spangrp 6 102 Error (Mask x Cond x Serpos) 455 Total Note: Spangrp = Span Group, Cond = Condition,

327

SS

MS

1.07 0.24 0.24 0.83 0.05 15.80 0.00 0.00 0.04 0.04 1.64 0.10 0.11 0.05 0.07 0.03 1.42 0.04 1.27 0.42 0.32 0.11 2.92 0.06 0.09 0.05 0.30 0.15 1.33 0.04 0.27 0.09 0.05 0.02 0.87 0.02 0.09 0.01 0.20 0.03 1.99 0.02 0.09 0.02 0.20 0.03 2.53 0.02 16.87 Serpos = Serial Position

F 4.82

0.03 0.41 1.28 0.82 7.39 1.87 1.19 3.88 5.29 0.93 0.73 1.68 0.62 1.31

Table B 25: Summary table for repeated measures A N O V A on naming errors by mask, task and condition in Experiment 2d Source Within Subjects Mask Error (Mask) Task Error (Task) Cond Error (Cond) Mask x Task Error (Mask x Task) Mask x Cond Error (Mask x Cond) Task x Cond Error (Task x Cond) Mask x Task x Cond Error (Mask x Task x Cond) Total

df 24

SS 0.11

1 24 1 24 2 48 1 24 2 48 2 48 2 48 299

MS 0.00 0.13 0.01 0.08 0.12 0.27 0.00 0.13 0.02 0.17 0.01 0.24 0.00 0.15

1.44

Note: Cond = Condition

328

F 0.00 0.01 0.01 0.00 0.06 0.01 0.00 0.01 0.01 0.00 0.01 0.00 0.00 0.00

0.26 1.79 10.85 0.21 3.12 1.05 0.66

Table B 26: Summary table for mixed design A N O V A on naming errors by span group, mask, task and condition in Experiment 2d Source SS df 18 0.09 Between Subjects 0.01 Spangrp 1 0.08 Error (Spangrp) 17 0.92 Within Subjects 209 0.00 Mask 1 1 0.00 Mask x Spangrp 0.08 17 Error (Mask) 0.01 1 Task 0.01 1 Task x Spangrp 0.05 Error (Task) 17 0.11 2 Cond 0.00 2 Cond x Spangrp 0.20 34 Error (Cond) 0.00 1 Mask x Task 0.00 1 Mask x Task x Spangrp 0.07 17 Error (Mask x Task) 0.01 2 Mask x Cond 0.00 2 Mask x Cond x Spangrp 0.12 34 Error (Mask x Cond) 0.00 2 Task x Cond 0.00 2 Task x Cond x Spangrp 0.15 34 Error (Task x Cond) 0.00 2 Mask x Task x Cond 0.00 2 Mask x Task x Cond x Spangrp 0.11 34 Error (Mask x Task x Cond) 1.01 227 Total Note: Spangrp = Span Group, Cond = Condition

329

F

MS 0.01 0.00

1.20

0.00 0.00 0.00 0.01 0.01 0.00 0.06 0.00 0.01 0.00 0.00 0.00 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

0.03 0.04 3.36 5.27 9.37 0.34 0.25 0.46 1.91 0.16 0.01 0.33 0.74 0.34

Table B 27: Summary table for repeated measures A N O V A on naming latencies by

mask, task, condition and serial position in Experiment 2d for aware participants Source Within Subjects Mask Error (Mask) Task Error (Task) Cond Error (Cond) Serpos Error (Serpos) Mask x Task Error (Mask x Task) Mask x Cond Error (Mask x Cond) Task x Cond Error (Task x Cond) Mask x Task x Cond Error (Mask x Task x Cond) Mask x Serpos Error (Mask x Serpos) Task x Serpos Error (Task x Serpos) Mask x Task x Serpos Error (Mask x Task x Serpos) Cond x Serpos Error (Cond x Serpos) Mask x Cond x Serpos Error (Mask x Cond x Serpos) Task x Cond x Serpos Error (Task x Cond x Serpos) Mask x Task x Cond x Serpos Error (Mask x Task x Cond x Serpos)

df 13 1 13 1 13 2 26 3 39 1 13 2 26 2 26 2 26 3 39 3 39 3 39 6 78 6 78 6 78 6 78 671

Total Note: Cond = Condition, Serpos = Serial Position

330

SS 123118.84 362653.75 309305.12 1807947.52 470025.73 225144.17 48762.13 208889.06 245277.06 45.05 78849.91 10391.93 55769.25 4536.14 39319.99 3403.50 33415.35 13989.70 94009.71 45662.52 213947.78 4854.30 57934.21 10945.90 113057.60 1561.97 88333.55 18718.13 132102.21 12960.51 114335.18 4949267.77

MS 362653.75 23792.70 1807947.52 36155.83 112572.08 1875.47 69629.69 6289.16 45.05 6065.38 5195.97 2144.97 2268.07 1512.31 1701.75 1285.21 4663.24 2410.51 15220.84 5485.84 1618.10 1485.49 1824.32 1449.46 260.33 1132.48 3119.69 1693.62 2160.08 1465.84

F 15.24 50.00 60.02 11.07 0.01 2.42 1.50 1.32 1.94 2.78 1.09 1.26 0.23 1.84 1.47

Appendix C Major statistical analyses performed in Experiment 3a and 3b

Table C 1: Summary table for repeated measures ANOVA on number corre condition and working memory load in Experiment 3 a Source Within Subjects Cond Error (Cond) Load Error (Load) Cond x Load Error (Cond x Load) Total

df

SS

45

294.42 66.98 136.39 2820.44 1515.68 48.64 497.49 5380.04

1 45 3 135 3 135 367

MS

F

66.98 3.03 940.15 11.23 16.21 3.69

22.10 83.74 4.40

Note: Cond = Condition, Load = Working Memory Load

Table C 2: Summary table for mixed design ANOVA on number correct by group, condition and working memory load in Experiment 3 a Source df SS MS F Between Subjects Spangrp Error (Spangrp) Within Subjects Cond Cond x Spangrp Error (Cond) Load Load x Spangrp Error (Load) Cond x Load Cond x Load x Spangrp Error (Cond x Load) Total

30 1 29 217 1 1 29 3 3 87 3 3 87 247

204.88 6.01 198.87 3388.31 40.68 1.00 96.99 1828.78 46.19 971.86 36.06 9.35 357.40 3593.19

6.01 6.86

0.88

40.68 1.00 3.35 609.59 15.40 11.17 12.02 3.12 4T1

12.16 0.30 54.57 1.38 2.93 0.76

Note: Spangrp = Span Group, Cond = Condition, Load = Working Memory Load

331

Table C 3: Summary table for repeated measures A N O V A on d' values by condition and working memory load in Experiment 3 a Source Within Subjects Cond Error (Cond) Load Error (Load) Cond x Load Error (Cond x Load) Total

df 45 1 45 3 135 3 135 367

SS 422.14 37.06 75.30 1011.05 422.10 5.56 276.64 2249.85

MS 37.06 1.67 337.02 3.13 1.85 2.05

F 22.15 107.79 0.90

Note: Cond = Condition, Load = Working Memory Load Table C 4: Summary table for mixed design ANOVA on d' values by span group, condition and working memory load in Experiment 3 a df SS MS Source 274.01 30 Between Subjects 1.03 9.42 9.42 1 Spangrp 9.12 264.59 29 Error (Spangrp) 1215.56 217 Within Subjects 18.31 32.08 32.08 1 Cond 1.28 2.24 2.24 1 Cond x Spangrp 1.75 50.82 29 Error (Cond) 67.88 215.76 647.27 3 Load 0.53 1.67 5.01 3 Load x Spangrp 3.18 276.52 87 Error (Load) 0.30 0.59 1.78 3 Cond x Load 4.38 8.74 26.23 3 Cond x Load x Spangrp 2.00 173.61 87 Error (Cond x Load) 247 1489.57 Total = Condition, Load = Working Memory Load Note: Spangrp = Span Group, Cond

332

Table C 5: Summary table for repeated measures A N O V A on response time by condition and working memory load in Experiment 3 a Source Within Subjects Cond Error (Cond) Load Error (Load) Cond x Load Error (Cond x Load) Total

df 45

1 45 3 135 3 135 367

SS

MS

1894695.78 171250.21 448443.44 9328270.15 5158118.45 72661.80 940488.82 18013929

F

171250.21 9965.41 3109423.38 38208.29 24220.60 6966.58

17.18 81.38 3.48

Note: Cond = Condition, Load = Working Memory Load Table C 6: Summary table for mixed design A N O V A on response time by span group, condition and working memory load in Experiment 3 a Source Between Subjects Spangrp Error (Spangrp) Within Subjects Cond Cond x Spangrp Error (Cond) Load Load x Spangrp Error (Load) Cond x Load Cond x Load x Spangrp Error (Cond x Load) Total

df

SS

30

1162361.87 120494.92 1041866.95 10510924.97 95487.38 12471.15 252520.83 6151515.11 28277.11 3251328.66 42433.17 35595.48 641296.08 11673286.84

1 29 217 1 1 29 3 3 87 3 3 87 247

MS

F

120494.92 35926.45

3.35

95487.38 12471.15 8707.62 2050505.04 9425.70 37371.59 14144.39 11865.16 7371.22

10.97 1.43 54.87 0.25 1.92 1.61

Note: Spangrp = Span Group, Cond = Condition, Load = Working Memory Load

333

Table C 7: Summary table for repeated measures A N O V A on number correct by condition and working memory load in Experiment 3 b Source Within Subjects Cond Error (Cond) Load Error (Load) Cond x Load Error (Cond x Load) Total

df Al

1 47 2 94 2 94 287

SS

MS

763.47 90.00 427.50 1808.17 1448.83 103.76 853.24 5494.97

F

90.00 9.10 904.09 15.41 51.88 9.08

9.90 58.66 5.72

Note: Cond = Condition, Load = Working Memory Load Table C 8: Summary table for mixed design ANOVA on number correct by span group, condition and working memory load in Experiment 3b Source Between Subjects Spangrp Error (Spangrp) Within Subjects Cond Cond x Spangrp Error (Cond) Load Load x Spangrp Error (Load) Cond x Load Cond x Load x Spangrp Error (Cond x Load) Total

df 39 1 38 200 1 1 38 2 2 76 2 2 16 239

SS 620.44 23.12 597.32 4134.48 113.88 3.21 373.25 1632.58 78.68 1105.56 90.09 26.56 710.67 4754.92

F

MS 23.12 15.72

1.47

113.88 3.21 9.82 816.29 39.34 14.55 45.05 13.28 9.35

11.59 0.33 56.11 2.70 4.82 1.42

Note: Spangrp = Span Group, Cond = Condition, Load = Working Memory Load

334

Table C 9: Summary table for repeated measures A N O V A on d' values by condition and working memory load in Experiment 3b Source Within Subjects Cond Error (Cond) Load Error (Load) Cond x Load Error (Cond x Load) Total

df Al

1 47 2 94 2 94 287

SS 234.74 11.89 58.60 185.46 247.11 14.82 86.34 838.96

MS

F 11.89 1.25 92.73 2.63 7.41 0.92

9.54 35.27 8.07

Note: Cond = Condition, Load = Working Memory Load Table C 10: Summary table for mixed design ANOVA on d' values by span group, condition and working memory load in Experiment 3b Source df SS MS F Between Subjects 39 158.97 Spangrp 1 2.53 2.53 0.62 Error (Spangrp) 38 156.44 4.12 Within Subjects 200 490.82 Cond 1 10.41 10.41 8.37 Cond x Spangrp 1 0.08 0.08 0.07 Error (Cond) 38 47.28 1.24 Load 2 140.96 70.48 27.01 Load x Spangrp 2 6.52 3.26 1.25 Error (Load) 76 198.35 2.61 Cond x Load 2 11.14 5.57 5.60 Cond x Load x Spangrp 0.49 0.24 2 0.25 Error (Cond x Load) 76 75.59 1.00 Total 239 649.79 Note: Spangrp = Span Group, Cond = Condition, Load = Working Memory Load

335

Table C 11: Summary table for repeated measures A N O V A on response time by condition and working memory load in Experiment 3b Source Within Subjects Cond Error (Cond) Load Error (Load) Cond x Load Error (Cond x Load) Total

df Al 1 47 2 94 2 94 287

SS 3488787.12 59233.67 845512.93 70145.37 1890677.12 2472.93 1097674.41 7454504

MS 59233.67 17989.64 35072.69 20113.59 1236.47 11677.39

F 3.29 1.74 0.11

Note: Cond = Condition, Load = Working Memory Load

Table C 12: Summary table for mixed design ANOVA on response time by span grou condition and working memory load in Experiment 3b Source F df SS MS Between Subjects 39 3167405.72 96489.84 96489.84 Spangrp 1.19 1 3070915.88 80813.58 Error (Spangrp) 38 Within Subjects 200 3227296.300 1.94 39522.57 39522.57 1 Cond 297.44 0.02 297.44 1 Cond x Spangrp 20384.81 774622.84 38 Error (Cond) 16543.17 0.95 33086.33 2 Load 1.72 29941.63 59883.26 2 Load x Spangrp 17461.36 1327063.56 76 Error (Load) 0.25 3167.70 6335.40 2 Cond x Load 0.68 8628.87 17257.74 2 Cond x Load x Spangrp 12752.99 969227.16 76 Error (Cond x Load) 6394702.02 239 Total Note: Soanero = Span Grout), Cond = Condition, Load = Working Memory Load

336